pmid,articleUrl,articleTitle,abstractText,authors,journalTitle,publicationDate,doi,publicationTypes,meshTerms,pmcId 42888301,https://pubmed.ncbi.nlm.nih.gov/42888301/,Synthetic Biology Tools and Strategies for Microbial Cell Factory Construction.,"Microbial cell factories have emerged as important platforms for the sustainable production of fuels, chemicals, pharmaceuticals, and other value-added bioproducts. Synthetic biology provides a systematic engineering framework for their construction and optimization through standardized biological parts, modular design, and iterative Design-Build-Test-Learn cycles. This chapter summarizes the major tools and strategies for developing microbial cell factories, beginning with the fundamental principles of synthetic biology and the characteristics of representative microbial chassis. Key genetic engineering approaches, including CRISPR-based genome editing, transcriptional and translational regulation, and modular DNA assembly, are discussed together with strategies for metabolic pathway design, enzyme engineering, flux balancing, and dynamic regulation. Genome-scale metabolic modeling, adaptive and directed evolution, and multi-omics analysis are further introduced as systems-level approaches for identifying engineering targets and improving cellular performance. In addition, emerging computational tools, machine learning, artificial intelligence, high-throughput screening, and laboratory automation are highlighted for their roles in accelerating strain development and closing the Design-Build-Test-Learn cycle. Finally, representative applications illustrate the translation of engineered microbial cell factories from laboratory-scale design to industrial biomanufacturing.","[""Jiang Jipeng"", ""Zhang Xin"", ""Zhang Lixin"", ""Tan Gao-Yi""]",Advances in biochemical engineering/biotechnology,2027-08-02,10.1007/10_2026_363,"[""Journal Article""]",, 42920311,https://pubmed.ncbi.nlm.nih.gov/42920311/,Modeling the Selective and Lossy Nature of Visual Processing Bridges the Discrepancy in Noninvasive Neural Decoding.,"Noninvasive neural decoding seeks to infer visual stimuli from brain recordings, such as EEG and MEG, thereby providing a computational approach for probing human visual information processing. Current contrastive learning approaches align brain signals with image embeddings from pretrained neural networks. However, these methods often struggle with the intrinsic visual-brain discrepancy arising from systematic representational gaps and noise because brain signals are treated as direct, lossless replicas of visual inputs. To alleviate this mismatch, the proposed framework accounts for the selective and information-compressive characteristics of biological vision. On the visual side, the framework incorporates an adaptive, context-aware visual calibration mechanism, where multilevel blurred representations are dynamically fused under the guidance of structural salience to refine visual features toward the brain's abstraction level. On the brain-signal side, a robust and lightweight neural encoder is used to mitigate inter-channel and temporal variations. Experiments on standard EEG and MEG benchmarks show that the proposed method achieves strong brain-to-image retrieval performance. Our modeling hypothesis, supported by retrieval behavior, suggests that computational representations aligning with salient and structurally meaningful content better match macroscopic neural responses than fine-grained image details. These findings suggest that incorporating the lossy and selective characteristics of human visual processing into computational models is important for advancing noninvasive neural decoding and understanding brain-machine alignment. The source codes are available at SLVP.","[""Li Yuanpeng"", ""Tan Xilin"", ""Tan Minjie"", ""Gao Shaobing""]",International journal of neural systems,2026-10-22,10.1142/S0129065727500298,"[""Journal Article""]","[""Humans"", ""Electroencephalography"", ""Magnetoencephalography"", ""Visual Perception"", ""Models, Neurological"", ""Brain"", ""Neural Networks, Computer""]", 42855827,https://pubmed.ncbi.nlm.nih.gov/42855827/,A Dual-Loop Task-Orthogonal Framework for Sequential Multitask Learning in Echo State Networks.,"Sequential multitask learning is a major problem for Echo State Networks (ESNs) because of catastrophic forgetting and interference between the tasks. To solve this, a Dual-Loop Task-Orthogonal Framework of Echo State Networks (DLTOF-ESN) model is proposed. It has a well-scaling architecture based on orthogonal parameter updates and a new mechanism named Adaptive Stability Regulation (ASR). DLTOF-ESN is based on dual-loop learning. The inner loop uses orthogonal updates to minimize gradient overlap. The outer loop adapts regularization strength based on changes in the relationships of tasks. Also, in this loop the weight sensitivity and decay-weighted similarity are measured using an adaptive relevance mask. This dual-loop interactive mechanism enables the related tasks to be flexible whilst maintaining important information in the tasks that have a temporal distance. In contrast to the traditional Elastic Weight Consolidation (EWC), which uses a fixed regularization, this DLTOF-ESN dynamically varies the penalties according to the relevance of the task and time. This research integrated SHAP-driven layers of interpretability to the framework, which enables explainability and is useful in high-stakes areas. Model evaluations are performed on three benchmark continual learning datasets, Permuted MNIST, Split CIFAR 100 and CIFAR-10, which have shown that DLTOF-ESN shows improvement in tasks recall of up to 18%, accuracy of up to 17%, and interference of up to 44% compared to EWC + ESN and OL + ESN architectures. Ablation studies confirm the synergy between orthogonal updates and ASR methods, while scalability tests reveal reduced training time and memory overhead with increasing dataset sizes.","[""Zareen Syeda Shamaila"", ""Wang Junsong"", ""Wang Hong"", ""Wang Wenting"", ""Wang Yi"", ""Liu Xiaoxu"", ""Kang Yan""]",International journal of neural systems,2026-10-10,10.1142/S0129065727500274,"[""Journal Article""]","[""Humans"", ""Neural Networks, Computer"", ""Machine Learning"", ""Adaptive Algorithms""]", 43014046,https://pubmed.ncbi.nlm.nih.gov/43014046/,"Stimuli-responsive (switchable) deep eutectic solvents in microextraction: Status, challenges and prospects in food safety and environmental analysis.","Deep eutectic solvents (DESs) have attracted considerable attention as green extraction solvents; however, their application in microextraction is often limited by difficult phase separation, poor recyclability, and the need for additional separation steps. Stimuli-responsive or switchable deep eutectic solvents (S-DES) overcome these limitations through reversible hydrophilic-hydrophobic transitions triggered by external stimuli such as CO2/N2, pH, or temperature, enabling efficient extraction, rapid phase separation, and solvent reuse. This review critically examines the current status, challenges, and prospects of S-DES in microextraction for food safety and environmental analysis. The design principles, classification, and switching mechanisms of CO2/N2-, pH-, temperature-, magnetic-, photo-, and electro-responsive S-DES are discussed and critically assessed, together with the molecular interactions governing analyte extraction. Recent applications in liquid-phase microextraction for pesticides, antibiotics, pharmaceuticals, mycotoxins, heavy metals, and emerging contaminants are systematically evaluated. The role of computational tools, including COSMO-RS, density functional theory, molecular dynamics simulations, and machine learning, in understanding extraction mechanisms and guiding solvent design is also highlighted. Current challenges include limited validation in complex matrices, a lack of standardized switching protocols, and the underutilization of temperature-responsive systems. Future research directions focusing on temperature-responsive extraction platforms, multi-class contaminant extraction, and AI-assisted solvent design are discussed. S-DES show strong potential for developing greener, simpler, and more efficient microextraction methods for food safety and environmental monitoring.","[""Javeria Huma"", ""Yahaya Noorfatimah"", ""Mohamed Ahmad Husaini"", ""Kamaruzaman Sazlinda"", ""Mohamad Zain Nur Nadhirah"", ""Herzog Grégoire"", ""Chen David"", ""Hong Lim Sheh""]",Talanta,2026-10-28,10.1016/j.talanta.2026.130863,"[""Journal Article"", ""Review""]","[""Liquid Phase Microextraction"", ""Deep Eutectic Solvents"", ""Food Safety"", ""Solvents"", ""Environmental Monitoring"", ""Temperature"", ""Hydrogen-Ion Concentration"", ""Hydrophobic and Hydrophilic Interactions""]", 43008372,https://pubmed.ncbi.nlm.nih.gov/43008372/,"Next-generation hyphenated analytical technologies in natural product research: From high-resolution separation and structural elucidation to dereplication, metabolomics and drug discovery.","Natural products occupy a unique position at the interface of chemistry, biology and medicine and continue to provide structurally diverse molecules for drug discovery, chemical biology, nutrition and therapeutic development. Their investigation, however, is complicated by the enormous chemical complexity of natural-product extracts, the presence of structurally related metabolites, isomeric and isobaric compounds, wide concentration ranges, matrix effects and the frequent occurrence of previously uncharacterized constituents. Over the past decade, advances in chromatographic separation, high-resolution mass spectrometry (HRMS), tandem mass spectrometry (MS/MS), ion mobility spectrometry (IMS), nuclear magnetic resonance (NMR) spectroscopy and computational analysis have fundamentally changed how these challenges are addressed. High performance liquid chromatography-mass spectrometry (HPLC-MS), HPLC-MS/MS, HPLC-NMR, LC-solid-phase extraction (SPE)-NMR and capillary electrophoresis-mass spectrometry (CE-MS) have evolved from individual hyphenated techniques into integrated analytical workflows capable of linking chemical separation with molecular characterization, dereplication and biological prioritization. In parallel, molecular networking, spectral databases, in silico fragmentation, machine learning and metabolomics platforms have transformed the interpretation of increasingly complex datasets. This review critically examines these developments, emphasizing the complementary information generated by chromatographic, MS and NMR measurements and their integration with computational approaches. Particular attention is given to HRMS/MS, data-independent acquisition, ion mobility, LC-SPE-NMR, molecular networking, spectral-library searching, metabolite annotation confidence and emerging artificial-intelligence-assisted workflows. The review argues that the principal transformation in natural-product analysis is not simply improved instrumental sensitivity, but the emergence of multidimensional and computationally integrated workflows that connect extraction, separation, detection, structural annotation, dereplication, metabolomics and bioactivity-guided discovery. Remaining challenges including incomplete databases, isomer discrimination, annotation uncertainty, reproducibility and the need for orthogonal structure confirmation are discussed, together with future opportunities for automated, multidimensional and FAIR-data-enabled natural-product discovery.","[""Dar Alamgir A"", ""Mukhtar Malik"", ""Shameem S A"", ""Singh Nasseb""]",Talanta,2026-11-05,10.1016/j.talanta.2026.130918,"[""Journal Article"", ""Review""]","[""Biological Products"", ""Metabolomics"", ""Drug Discovery"", ""Liquid Chromatography-Mass Spectrometry"", ""Magnetic Resonance Spectroscopy"", ""Chromatography, High Pressure Liquid"", ""Tandem Mass Spectrometry"", ""Mass Spectrometry""]", 42994314,https://pubmed.ncbi.nlm.nih.gov/42994314/,A novel deep learning-based peak detection technique for in vivo metabolite analysis of triterpenoids from Eleutherococcus senticosus leaves in Parkinson's disease mice.,"To address the challenges in mass spectrometry data processing, including poor detection of low-abundance metabolites, severe noise interference, and the reliance on labor-intensive parameter tuning in conventional peak detection methods, a deep learning-based peak detection algorithm termed MSPeakNet was developed. The model used a multi-scale convolutional module and a U-net encoder-decoder architecture integrated with an attention mechanism, enabling high-precision peak identification and noise reduction in complex biological matrices. MSPeakNet achieved precision and F1 scores comparable to CWT, with superior noise robustness, while maintaining competitive recall. The developed method was subsequently applied to the in vivo metabolite profiling of total triterpenoids from Eleutherococcus senticosus leaves (ESLT) in an MPTP-induced mouse model of Parkinson's disease (PD) using UPLC-Q-TOF/MS analysis. A total of 143 ESLT-related compounds, comprising 88 prototypes and 55 metabolites, were identified in plasma, urine, feces, and brain tissue, of which 16 prototypes were detected in brain tissue. Pharmacodynamic evaluation further confirmed that ESLT significantly ameliorated MPTP-induced motor dysfunction and dopaminergic neuronal loss. Comparative analysis further revealed distinct profiles of ESLT-related differential compounds between PD model and normal mice across multiple biological matrices; 14, 9, 6, and 4 differential metabolites were identified in feces, urine, plasma, and brain, respectively, suggesting that disease status may affect the biotransformation, distribution, and elimination of ESLT constituents. A robust and efficient deep learning-based approach for mass spectrometry peak detection was thereby established, and critical experimental evidence was provided for the discovery of anti-Parkinsonian active ingredients and quality markers from E. senticosus leaves.","[""Wen Tiantian"", ""Guan Wei"", ""Xu Shiqi"", ""Yang Xu"", ""Li Yanying"", ""Wang Hao"", ""Jiang Peng"", ""Chen Qingshan"", ""Zhang Lili"", ""Wang Quanyong"", ""Kuang Haixue"", ""Liu Shu"", ""Liu Yan""]",Talanta,2026-11-04,10.1016/j.talanta.2026.130904,"[""Journal Article""]","[""Animals"", ""Eleutherococcus"", ""Deep Learning"", ""Mice"", ""Triterpenes"", ""Plant Leaves"", ""Male"", ""Parkinson Disease"", ""Disease Models, Animal"", ""Plant Extracts"", ""Mice, Inbred C57BL"", ""Mass Spectrometry""]", 42987522,https://pubmed.ncbi.nlm.nih.gov/42987522/,Enrichment characteristics and hyperspectral quantitative retrieval of lithium and rare earth element concentrations in coal.,"Coal is a host of critical metals, and rapid, non-destructive quantification of these metals is important for resource assessment. This study investigated lithium (Li) and rare earth elements (REE) in coal samples from the Sihe Coal Mine, Shanxi Province, and the Hebi Coal Mine, Henan Province, China. Hyperspectral data (350-2500 nm) were processed using First-order Derivative (FD) and Continuum Removal (CR), followed by feature selection with Competitive Adaptive Reweighted Sampling (CARS) and Boruta. Partial Least Squares Regression (PLSR), Back-Propagation Neural Networks (BPNN), and Particle Swarm Optimization-Support Vector Regression (PSO-SVR) were used to develop quantitative retrieval models. Lithium was relatively enriched, whereas REE concentration were generally within the normal range. The FD-Boruta-BPNN model achieved the highest prediction accuracy for both Li (R2=0.912) and REE (R2=0.968), using fewer input wavelengths. These findings demonstrate that spectral preprocessing, feature selection, and modeling algorithms jointly influence retrieval performance, highlighting the need to balance predictive accuracy and model complexity. Predictions likely rely on indirect spectral information associated with mineral carriers rather than direct elemental absorption features. However, the limited sample size and geographical coverage constrain generalizability. Future research should incorporate larger, independent datasets from diverse geological settings, verify spectral-response mechanisms through mineralogical analysis, and extend the framework to other coal-hosted critical metals and hyperspectral imaging.","[""Zhao Hengqian"", ""Cheng Sijun"", ""Wang Mengmeng"", ""Zhao Lei"", ""Li Chige"", ""Sun Shasha"", ""Wei Yingchun"", ""Tian Yufei""]",Talanta,2026-10-26,10.1016/j.talanta.2026.130834,"[""Journal Article""]",, 42967395,https://pubmed.ncbi.nlm.nih.gov/42967395/,Colorimetric sensor arrays and artificial intelligence for natural product quality evaluation: From cross-reactive fingerprints to analytical validation.,"Natural products encompass traditional Chinese medicines, herbal medicines, teas, food-derived plant materials, plant extracts, and naturally occurring bioactive compounds. Their rapid identification and quality evaluation require a balance among integrative characterization, analytical accuracy, and field accessibility. Empirical authentication methods are difficult to quantify and reproduce, whereas confirmatory techniques based on chromatography, mass spectrometry, and spectroscopy, although analytically reliable, cannot fully meet the frequent demands of on-site screening, process monitoring, and risk triage. A colorimetric sensor array (CSA) generates sample-specific color fingerprints through cross-reactive responses across multiple sensing elements. Combined with standardized image acquisition, feature extraction, and chemometric or machine-learning analysis, CSA-based workflows can support classification, quantitative prediction, and anomaly detection. This integrated strategy is referred to here as CSA-AI, an umbrella term that does not equate conventional chemometrics with artificial intelligence (AI). This review critically examines CSA design, color-data acquisition, chemometric and AI-assisted analysis, analytical validation, and applications in natural product quality evaluation along the chain of ""array response-color fingerprint-computational interpretation-quality decision-making."" Particular attention is given to authentic complex matrices, sample-level data partitioning, data leakage, independently sourced external evaluation, cross-device transfer, validation evidence, model interpretability, and confirmation by established analytical methods. Current direct visible-CSA evidence is strongest for authentication and preliminary adulteration screening, whereas process monitoring, storage-deterioration warning, and field deployment remain less thoroughly validated. Current evidence supports CSA-AI primarily as a complementary tool for rapid screening, risk warning, and auxiliary grading rather than as a replacement for confirmatory methods such as HPLC, GC-MS, and LC-MS.","[""Chen Jie-Feng"", ""Yi Wan-Yu"", ""Huang Yu-Ting"", ""Huang Yan"", ""Lin Wei-Wen"", ""Chen Jie-Xin"", ""Qian Zheng-Ming"", ""Ma Hong-Yan"", ""Tang Dan""]",Talanta,2026-10-28,10.1016/j.talanta.2026.130854,"[""Journal Article"", ""Review""]","[""Biological Products"", ""Artificial Intelligence"", ""Colorimetry"", ""Quality Control""]", 42966235,https://pubmed.ncbi.nlm.nih.gov/42966235/,"Thyroid biomarker biosensing: Nanomaterial-enabled electrochemical platforms, clinical translation, point-of-care technologies, and intelligent diagnostics.","Thyroid disorders represent a substantial clinical burden, requiring reliable biomarker measurement for diagnosis, monitoring, and disease assessment. Thyroid-stimulating hormone (TSH), triiodothyronine (T3), thyroxine (T4), thyroglobulin (Tg), calcitonin (CT), and thyroid-associated nucleic acid biomarkers are important analytical targets. Although established immunoassays, ELISA, imaging, and cytological methods remain integral to clinical practice, limitations related to laboratory infrastructure, sample requirements, and workflow complexity have driven the development of alternative analytical platforms. This review focuses primarily on nanomaterial-assisted electrochemical biosensors for thyroid biomarker analysis, examining how nanomaterial properties, biorecognition strategies, and electrochemical transduction influence analytical performance. Metal and metal oxide nanoparticles, carbon nanomaterials, metal-organic frameworks, metal chalcogenides, MXenes, quantum dots, and molecularly imprinted polymers are discussed for their roles in signal amplification, interfacial charge transfer, selectivity, and sensor stability. Complementary approaches, including chemiluminescent, photoelectrochemical, optical, and other emerging sensing strategies, are also considered to contextualize the broader analytical landscape, particularly in relation to their implementation in commercial laboratory and point-of-care platforms for thyroid analysis. Seventy representative analytical and diagnostic systems are comparatively evaluated according to sensing technology, detection range, sample type, and assay time, together with their progression toward portable point-of-care devices, commercial analyzers, rapid diagnostic formats, and wearable platforms. Emerging artificial intelligence and machine learning approaches for signal interpretation, disease classification, and risk assessment are discussed. Challenges in nanomaterial reproducibility, bioreceptor stability, biofouling, matrix effects, fabrication, analytical standardization, external validation, and clinical translation are critically examined. The review identifies key analytical and technological priorities for translating thyroid biosensors from sensitive experimental platforms into robust, reproducible, and clinically relevant diagnostic tools.","[""Bathinapatla Ayyappa"", ""Pakrudheen Iqbal"", ""Kanchi Suvardhan""]",Talanta,2026-10-28,10.1016/j.talanta.2026.130867,"[""Journal Article"", ""Review""]","[""Humans"", ""Biosensing Techniques"", ""Biomarkers"", ""Electrochemical Techniques"", ""Point-of-Care Systems"", ""Nanostructures"", ""Thyroid Diseases"", ""Thyroid Gland"", ""Translational Research, Biomedical""]", 42956610,https://pubmed.ncbi.nlm.nih.gov/42956610/,Deep learning-assisted portable sensing platform based on a fluorescent MOF-polymer mixed-matrix membrane for rapid detection of nitrazepam in biological samples.,"Nitrazepam (NZP) is a benzodiazepine sedative-hypnotic widely used in the treatment of anxiety and sleep disorders. However, non-medical use of NZP as a new psychoactive substance poses serious health risks. Therefore, rapid and accurate detection of NZP in biological samples is important for point-of-care testing. Herein, a deep learning-assisted fluorescent sensing platform was developed based on a mixed-matrix membrane (1@PMMA), fabricated by incorporating a Cd-MOF, {[Cd(L)(H2O)2]·3H2O}n (1) (H2L = 5-(3-(pyridin-4-yl)ureido)isophthalic acid), into a poly(methyl methacrylate) (PMMA) matrix for NZP detection. The resulting 1@PMMA exhibited good mechanical stability, processability, and fluorescence stability in biological samples. Upon exposure to NZP, 1@PMMA showed efficient fluorescence quenching, with a quenching efficiency of 96.19%, a detection limit of 0.21 μM, and a response time of approximately 2 s. The membrane maintained high selectivity in the presence of common biological components and stable sensing performance over five quenching-regeneration cycles. A smartphone-based fluorescence sensing method enabled portable visual detection of NZP through extraction of RGB values from fluorescence images of 1@PMMA. Additionally, a deep learning-based photo-to-result platform employing a ResNet-18 model directly predicted NZP concentrations from fluorescence images without manual data processing, showing a strong linear relationship between predicted and experimental concentrations (R2 = 0.9949). For spiked urine and serum samples, recoveries of 95.18-100.46% were obtained, with relative standard deviations of 0.18-3.10%. This work integrates a fluorescent MOF-based mixed-matrix membrane with a deep learning-assisted photo-to-result strategy for on-site quantitative determination of NZP in biological samples.","[""Lei Mingyuan"", ""Yang Kunnan"", ""Ma Junchao"", ""Zhang Yanbin"", ""Lu Wenhui"", ""Li Xuebo"", ""Huo Sihan"", ""Pei Haimeng"", ""Zhang Qingfu""]",Talanta,2026-10-22,10.1016/j.talanta.2026.130815,"[""Journal Article""]","[""Deep Learning"", ""Humans"", ""Nitrazepam"", ""Limit of Detection"", ""Spectrometry, Fluorescence"", ""Metal-Organic Frameworks"", ""Polymethyl Methacrylate"", ""Fluorescent Chemosensor Compounds"", ""Fluorescent Dyes"", ""Fluorescence""]", 42950637,https://pubmed.ncbi.nlm.nih.gov/42950637/,Improved SERS detection of tetrahydrocannabinol using 4-aminophenol derivatization and machine learning.,"Fast, sensitive, and selective detection of Δ9-tetrahydrocannabinol (THC) is essential for forensic screening and quality control of cannabis-derived products. Although surface-enhanced Raman spectroscopy (SERS) is promising for THC detection, its practical use remains limited by insufficient sensitivity and selectivity, especially in complex cannabis-infused samples. In this work, chemical derivatization was used to enhance SERS performance for THC detection. Two commonly used colorimetric reagents, i.e., Fast Blue BB (FBBB) and 4-aminophenol (4-AP), were evaluated with silver nanorod SERS substrates. Among the tested reagents, 4-AP showed better compatibility with SERS, yielding clear and enhanced Raman fingerprints at multiple bands, with 1100 cm-1 as the most suitable THC marker. Compared with our previous non-derivatized approach, the developed method lowered the detection limit by 20-fold, reaching 0.051 mg/L. The method also enabled spectral differentiation of THC in the presence of other prominent cannabinoids, including cannabidiol (CBD), cannabinol (CBN) and tetrahydrocannabinolic acid (THCA). Its applicability was further demonstrated by detecting THC in commercial cannabis-infused beverage samples, achieving a sub-ppm detection limit without sample extraction. Multivariate PCA and supervised machine learning models were further developed to support data analysis. Using selected THC-related bands improved chemical interpretability and provided more consistent performance across the evaluated models. Classification accuracies were 92.5-98.7% for discrimination between THC-containing and THC-free samples and 84.8-88.4% for threshold-based screening of spiked beverage samples. Overall, combining simple chemical derivatization with SERS and machine learning provides a practical approach for sensitive and selective THC analysis, with potential for screening cannabis-infused products.","[""Thepin Nopphamat"", ""Somboonsaksri Pacharamon"", ""Lapkrajang Davis"", ""Termsaithong Nichakarn"", ""Limwichean Saksorn"", ""Jintamethasawat Rungroj"", ""Botta Raju"", ""Ngernsutivorakul Thitaphat""]",Talanta,2026-10-13,10.1016/j.talanta.2026.130735,"[""Journal Article""]","[""Dronabinol"", ""Spectrum Analysis, Raman"", ""Aminophenols"", ""Machine Learning"", ""Limit of Detection"", ""Silver"", ""Nanotubes""]", 42855826,https://pubmed.ncbi.nlm.nih.gov/42855826/,Prototype-Guided Test-Time Adaptation for Real-Time EMG-Based Gesture Recognition.,"Surface electromyography (sEMG) signals facilitate intuitive human-machine interaction and are widely employed in applications such as prosthetics and rehabilitation. However, sEMG signal distributions are highly sensitive to variations in electrode placement, user physiology, and session conditions, which significantly hinder model generalization in real-world settings. To address this, we propose EMG-based structure-aware test-time adaptive refinement (EMG-STAR), a test-time adaptation (TTA) framework tailored for online deployment without access to target data during training. EMG-STAR improves model stability by fusing historical and batch-wise normalization statistics from both source and target domains, effectively alleviating domain shifts. Furthermore, it refines predictions using a prototype-assisted feature clustering method that enforces consistency among semantically similar target samples. To further enhance reliability, the framework jointly leverages prediction confidence and distributional alignment to filter out noisy pseudo-labels during adaptation. Comprehensive evaluations conducted on two public datasets, Hyser, and NinaPro-DB2, demonstrate that EMG-STAR consistently outperforms ten existing TTA methods. Interpretability analysis using Grad-CAM reveals that EMG-STAR predominantly attends to physiologically meaningful muscle activation regions, suggesting that the model captures neuromuscular structures rather than transient signal noise.","[""Ma Yingtong"", ""Wang Yalin"", ""Dai Chenyun"", ""Guo Yao""]",International journal of neural systems,2026-10-10,10.1142/S0129065727500249,"[""Journal Article""]","[""Humans"", ""Electromyography"", ""Gestures"", ""Pattern Recognition, Automated"", ""Adaptive Algorithms"", ""Signal Processing, Computer-Assisted"", ""Muscle, Skeletal""]", 42715211,https://pubmed.ncbi.nlm.nih.gov/42715211/,Neural Stability Enhancement Through Dynamic Lipschitz Constraints in Generative Adversarial Networks.,"Generative adversarial networks (GANs) are widely used for generative modeling in neural networks, yet their practical use is hindered by training instability and mode collapse. Existing stabilization approaches predominantly rely on enforcing fixed 1-Lipschitz constraints on the discriminator through techniques such as gradient penalty, spectral normalization, and gradient normalization. However, these methods fail to account for the dynamic nature of neural network training, where the appropriate Lipschitz constant target may vary with evolving system dynamics. This work challenges the conventional fixed-constraint paradigm by formalizing how the effective constraint strength and a bound-implied Lipschitz constant target can depend on architecture and training state. We provide theoretical analysis to motivate adaptive Lipschitz constraint (ALC) control and introduce ALC, a modular mechanism that adjusts the Lipschitz constant target rather than introducing another standalone regularization term. Experiments across multiple stable GAN architectures show improvements in generation quality and training stability in several evaluated settings. Our contributions advance the understanding of neural system stability in adversarial learning and provide a practical approach to ALC control in adversarial objectives.","[""Zhang Juexin"", ""Weng Ying"", ""Zhao Xueping"", ""He Xiangjian""]",International journal of neural systems,2026-09-08,10.1142/S0129065727500195,"[""Journal Article""]","[""Generative Adversarial Networks"", ""Neural Networks, Computer"", ""Algorithms"", ""Generative Artificial Intelligence"", ""Humans""]", 42428234,https://pubmed.ncbi.nlm.nih.gov/42428234/,Evaluation of Large Language Models in the Clinical Management of Patients With Upper Gastrointestinal Bleeding: Insights From Real-World Patient Data.,"OBJECTIVE: To evaluate large language models (LLMs) for pre-endoscopy (PE) risk stratification and prediction of endoscopic findings in upper gastrointestinal bleeding (UGIB), and compare their performance with clinical risk scores, conventional machine learning (ML) models, and hybrid approaches. METHODS: This multicenter retrospective study included 384 patients with UGIB who underwent endoscopy. Five LLMs (GPT-5, Gemini-2.5-Flash, Llama 4, Grok, and DeepSeek R1) were tested using structured zero-shot prompts based on PE clinical and laboratory data. Their performance in identifying high-risk patients was compared with the Glasgow-Blatchford Score (GBS), AIMS65, PE Rockall score, and ML models. Hybrid LLM-score models were also assessed. Two gastroenterologists evaluated the quality of LLM-generated justifications. RESULTS: GBS showed the best discriminative performance among clinical scores (area under the receiver operating characteristic curve [AUROC] 0.73). Among LLMs, GPT-5 achieved the highest accuracy (0.66), while Grok showed the best-balanced performance (0.59; F1 0.47). Gemini-2.5-Flash had the highest sensitivity (0.89) but low specificity (0.21). Endoscopic prediction performance was modest, with Gemini-2.5-Flash achieving the highest exact-match accuracy (0.34) and micro-F1 (0.38). Hybrid models improved performance over standalone LLMs but did not outperform GBS alone (best: GBS+GPT-5, AUROC 0.670. LLMs showed higher numerical performance than conventional ML models, but a statistical comparison was not possible due to unavailable instance-level data. Grok received the highest human evaluation score for explanation quality. CONCLUSIONS: LLMs showed moderate performance in UGIB risk stratification and endoscopic prediction but were inferior to clinical scores, especially GBS. Hybrid models modestly improved over standalone LLMs but not GBS, supporting their use as adjunct tools rather than clinical decision-support systems. TRIAL REGISTRATION: N/A.","[""Rajabnia Mohsen"", ""Davoodi Farbod"", ""Hajisafarali Mahsa"", ""Asadinia Shadi"", ""Shiri Mahsa"", ""Shirzad Fatemeh"", ""Saeedi Peyman"", ""Mohammadi Mahsa""]",DEN open,2026-07-09,10.1002/deo2.70373,"[""Journal Article""]",,PMC13347626 43061468,https://pubmed.ncbi.nlm.nih.gov/43061468/,4P-model: A machine learning-driven approach to expand non-invasive prenatal testing for early risk stratification of placenta accreta spectrum via genome-wide cfDNA promoter profiling.,"Placenta accreta spectrum (PAS) is a leading cause of life-threatening obstetric hemorrhage, yet early diagnosis remains challenging. Traditional non-invasive prenatal testing (NIPT) is limited to fetal aneuploidy screening, while its potential for maternal placental disorders is underexplored. We conducted a multicenter nested case-control study to investigate whether machine learning-driven genome-wide cell-free DNA (cfDNA) promoter profiling, derived from routine NIPT data, could extend the utility of NIPT to enable early prediction and subtype risk stratification of PAS. NIPT data were analyzed from 1060 individuals with a history of uterine surgery (126 increta/percreta, 139 accreta, and 795 controls after propensity score matching). A training set of 280 participants was used for feature selection with recursive feature elimination across five machine learning algorithms, and the best-performing model was validated in an internal cohort, a PLUS cohort with deeper sequencing, and two external cohorts generated on independent platforms. Pairwise differential analyses identified 211 candidate genes with altered promoter coverage. After feature selection, we developed the 4P-model, a 15-gene multiclass risk stratification model, to differentiate increta/percreta, accreta, and controls with a macro-average AUC of 0.973. Robust performance was reproduced across three NIPT platforms and the PLUS cohort, with AUCs of 0.882-0.946. These findings support that machine learning facilitates the translation of genome-wide cfDNA promoter signals embedded in NIPT data into early prediction and risk stratification of PAS. This noninvasive and scalable approach extends the diagnostic window, offering a clinically feasible tool for timely referral, improving perinatal outcomes.","[""Hao Runrun"", ""Wang Beilun"", ""Zhang Bin"", ""Li Yixuan"", ""Zhu Qingwen"", ""Zhu Shilin"", ""Liu Weiling"", ""Wu Yihan"", ""Zhang Lina"", ""Bai Xueqi"", ""Wang Shanshan"", ""Hu Shiman"", ""Zhang Yuting"", ""Shi Haoyan"", ""Liang Dong"", ""Yang Huixia"", ""Shi Zhonghua""]",Genes & diseases,2026-06-13,10.1016/j.gendis.2026.102305,"[""Journal Article""]",,PMC13783229 42959965,https://pubmed.ncbi.nlm.nih.gov/42959965/,Integrating single-cell transcriptomics and machine learning to reveal prognostic and immune significance of lactylation in breast cancer.,,"[""Qian Da"", ""Liu Haotian"", ""Mou Jianchen"", ""Yang Zhuotao"", ""Zhu Liquan"", ""He Chaoqi"", ""Wang Bo"", ""Meng Xuli""]",Genes & diseases,2026-02-16,10.1016/j.gendis.2026.102099,"[""Journal Article""]",,PMC13718854 42901643,https://pubmed.ncbi.nlm.nih.gov/42901643/,Discriminatory ability of WL Nutrient Agar for yeasts associated with olive environments: a binary scoring and machine learning approach.,"Wallerstein Laboratory Nutrient Agar (WLNA) is widely used to differentiate yeast colonies in wine and brewing microbiology, but its discriminatory ability for species associated with table olives and olive oil production environments has never been formally evaluated. We assessed colony features of 227 ascomycetous yeast isolates from these matrices - representing twelve species, predominantly Pichia membranifaciens, P. manshurica, P. kudriavzevii, Wickerhamomyces anomalus, Saccharomyces cerevisiae, Kluyveromyces marxianus, and Candida boidinii - using a binary scoring scheme applied to thirteen characters related to colony colour, shape, appearance and odour and four related to cell shape. Species identities were confirmed by MALDI-TOF mass spectrometry and ITS sequencing (overall agreement: accuracy = 0.90). Initial machine learning classification (CART, Random Forest, SVM-linear, SVM-RBF) on colony features yielded modest performance (accuracy 0.51-0.65), with reliable discrimination limited to S. cerevisiae and Nakazawaea molendinolei. A reproducibility study involving ten operators of varying skill levels identified the root causes: colour and elevation characters contributed disproportionately to inter-operator disagreement while Pichia spp. showed inherently high intraspecific variability. Revised models excluding characters contributing most to disagreement and pooling Pichia species at genus level achieved improved performance (Random Forest: accuracy = 0.72), with acceptable classification of S. cerevisiae, N. molendiolei, and Pichia spp. These results demonstrate that WLNA provides reasonable accuracy in the discrimination of some key species when using robust morphological characters, supporting its use as a low-cost screening tool prior to molecular identification, and establishes a methodological framework that is directly transferable to the evaluation of other differential media.","[""Parente Eugenio"", ""Pietrafesa Rocchina"", ""Romano Vito Antonio"", ""Labella Maria Grazia"", ""Ricciardi Annamaria""]",Food microbiology,2026-08-24,10.1016/j.fm.2026.105282,"[""Journal Article"", ""Evaluation Study""]","[""Olea"", ""Yeasts"", ""Machine Learning"", ""Culture Media"", ""Agar"", ""Pichia"", ""Mycological Typing Techniques"", ""Random Forest"", ""Saccharomyces cerevisiae"", ""Food Microbiology""]", 42901061,https://pubmed.ncbi.nlm.nih.gov/42901061/,Fe-porphyrin MOF nanozyme-based colorimetric sensor assisted by machine learning for detecting antioxidants in fruit juices.,"Antioxidants (AOs) are key bioactive components governing the nutritional quality and oxidative stability of fruit juices. However, their similar reducing properties make it challenging for conventional single-signal colorimetric methods to combine individual quantitative analysis with pattern recognition in complex food matrices. Herein, an Fe-porphyrin metal-organic framework (FeP-MOF) nanozyme was constructed using iron(III) meso-tetra(4-carboxyphenyl)porphyrin chloride (Fe-TCPPCl) as the functional ligand, and further integrated into a platform for high-throughput quantification and machine-learning-assisted pattern recognition. The porous MOF framework enhanced the accessibility and catalytic efficiency of Fe-porphyrin active sites. Based on AOs-mediated inhibition of the FeP-MOF/H2O2/TMB oxidation reaction, a microplate-based colorimetric sensor was developed for quantitative detection of ascorbic acid, glutathione, cysteine, gallic acid, and tannic acid, with limits of detection of 0.52-41.54 μM. Moreover, a multi-pH cross-reactive sensor array converted single absorbance signals into composition-sensitive multidimensional fingerprints. Coupled with linear discriminant analysis and hierarchical cluster analysis, the array achieved cross-validated classification accuracies of 100.0% for five antioxidants and 98.0% for five fruit-juice categories. Results also agreed well with high-performance liquid chromatography measurements. This work integrates nanozyme amplification, quantitative analysis of individual antioxidants, fingerprint sensing, and data-driven classification, providing a versatile strategy for high-throughput AOs analysis and compositional evaluation in complex food systems.","[""Kong Yijie"", ""Feng Xinyu"", ""Su Xindan"", ""Bi Yanming"", ""Zhang Lu"", ""Liu Sha""]",Talanta,2026-10-19,10.1016/j.talanta.2026.130805,"[""Journal Article""]","[""Machine Learning"", ""Colorimetry"", ""Antioxidants"", ""Fruit and Vegetable Juices"", ""Metal-Organic Frameworks"", ""Ascorbic Acid"", ""Iron"", ""Tannins"", ""Gallic Acid"", ""Glutathione"", ""Cysteine"", ""Metalloporphyrins"", ""Limit of Detection""]", 42896353,https://pubmed.ncbi.nlm.nih.gov/42896353/,Molecular arrangements of Thionine within the DNA scaffolds enable machine learning-assisted colorimetric LAMP detection.,"In the present work, we describe the distinctive molecular arrangement behavior of the phenothiazine dye - thionine (Th) - in the presence of LAMP amplicons and demonstrate its potential use as a visual indicator for LAMP-based detection. To better understand the mechanism responsible for DNA-templated stacking of Th molecules, detailed photophysical investigations were carried out in the presence of DNA and poly(acrylic acid) (PAA). Molecular docking analysis using AutoDock suggested that the planar phenothiazine aromatic core of Th is positioned along the groove region with the -NH2 groups oriented toward the outer region of the groove, remaining partially exposed to the solvent. Taking advantage of this interaction behavior, Th was further utilized to develop a colorimetric assay for nucleic acid amplification using LAMP. The dye exhibited a distinct color transition from violet to blue in the presence of LAMP amplicons generated by targeting the pneumolysin (ply) gene of Streptococcus pneumoniae. This visual change was accompanied by a bathochromic shift in the absorption spectrum, indicating DNA-induced molecular arrangements of Th molecules. The developed Th-based LAMP assay showed a limit of detection of 0.1 pg/μL. In addition, a machine learning-assisted image analysis approach was explored to quantitatively evaluate the colorimetric responses under standardized imaging conditions.","[""Sharma Arumugam Selva"", ""Yuk Ah Yun"", ""Lee Hyunji"", ""Lee Nae Yoon""]",Talanta,2026-10-14,10.1016/j.talanta.2026.130789,"[""Journal Article""]","[""Phenothiazines"", ""Colorimetry"", ""Machine Learning"", ""Nucleic Acid Amplification Techniques"", ""DNA"", ""Molecular Docking Simulation"", ""Acrylic Resins"", ""Streptococcus pneumoniae""]", 42892017,https://pubmed.ncbi.nlm.nih.gov/42892017/,Chemical discrimination of wild and cultivated Dendrobium devonianum by ATR-FTIR spectroscopy and machine learning with targeted metabolomic analysis.,"Wild and cultivated Dendrobium devonianum are visually similar, but their chemical differences complicate raw-material authentication. Recent spectroscopic studies of Dendrobium have focused primarily on species identification and geographical traceability, with limited chemical interpretation of wild-cultivated discrimination. ATR-FTIR spectroscopy and machine learning were applied to 145 samples (74 wild and 71 cultivated), while targeted metabolomics was performed using six pooled composite samples (three independently prepared pools per group). Spectral differences were concentrated in the 1800-900 cm-1 fingerprint region, although unsupervised analyses showed substantial overlap between groups. Three of 36 preprocessing-model combinations achieved 100% accuracy in an initial 43-sample test set. However, repeated stratified five-fold cross-validation (10 repeats) yielded a highest mean balanced accuracy of 0.9522 ± 0.0496 for partial least-squares discriminant analysis of raw spectra, providing a more cautious assessment of discrimination performance. SHAP identified 1058-1066 cm-1 as a key region associated with carbohydrate and glycosidic vibrations. Targeted metabolomics revealed 22 differential metabolites spanning sugars, amino acid-related compounds, and flavonoids. Among them, d-mannose, maltose, glucose, l-rhamnose, and glycosylated flavonoids showed associations with discriminative infrared variables. Compared with classification-focused studies, this framework connects spectral discrimination with complementary metabolite-level evidence, supporting more interpretable source authentication while highlighting the need for independent external validation.","[""Zheng Chuanmao"", ""Wang Yuanzhong""]",Talanta,2026-10-15,10.1016/j.talanta.2026.130792,"[""Journal Article""]","[""Dendrobium"", ""Spectroscopy, Fourier Transform Infrared"", ""Machine Learning"", ""Metabolomics"", ""Discriminant Analysis"", ""Flavonoids""]", 42892010,https://pubmed.ncbi.nlm.nih.gov/42892010/,Aptamer-CRISPR/Cas as a tool in point-of-care testing of infectious diseases.,"Aptamers and the CRISPR/Cas system have become the cornerstones of target detections due to their numerous advantages. Aptamers enable direct recognition of non-nucleic acid targets, such as pathogens and antigens, while CRISPR/Cas system drives signal amplification and rapid detection. The combined aptamer-CRISPR/Cas system can overcome the limitations of traditional target detection by realizing field-deployable, highly sensitive, specific, versatile, and rapid assays. Given the increasing number of different aptamer-CRISPR/Cas systems developed recently, particularly for infectious diseases, we provide a comprehensive overview of these platforms designed for detecting significant viruses such as SARS-CoV-2, Monkeypox virus, Hepatitis B virus and bacteria that include Mycobacterium tuberculosis, Staphylococcus aureus and Salmonella. Current challenges that impede the deployment of aptamer-CRISPR/Cas for point-of-care pathogen detection were also mapped, including primary barriers such as clinical matrix effect and stability. Several strategies to mitigate these issues were also set forth, predominantly the integration of artificial intelligence and machine learning to accelerate the transition of aptamer-CRISPR/Cas system to the clinical translation and commercialization phase, paving the way towards widespread implementation in global health.","[""Lau Cia-Hin"", ""Huang Yumei"", ""Ch'ng Ewe Seng"", ""Tang Thean-Hock"", ""Zhu Haibao"", ""Citartan Marimuthu""]",Talanta,2026-10-10,10.1016/j.talanta.2026.130767,"[""Journal Article"", ""Review""]","[""Humans"", ""CRISPR-Cas Systems"", ""Aptamers, Nucleotide"", ""Point-of-Care Testing"", ""Communicable Diseases"", ""Rapid Diagnostic Tests""]", 42880027,https://pubmed.ncbi.nlm.nih.gov/42880027/,Deep learning-enabled inertial microfluidic biosensor for high-efficiency automated point-of-care blood cell counting.,"While rapid quantification of blood cells is fundamental to modern hematological diagnosis and treatment, traditional hematology analyzers are constrained by bulky, costly hardware and the need for skilled operators, restricting their practical deployment in point-of-care testing (POCT) and primary healthcare settings. This work successfully develops a high-precision, low-cost microfluidic automated sensing platform for point-of-care cell quantitative counting by integrating inertial microfluidic technology with an image recognition algorithm. By introducing a novel loss function for small targets, integrating a multi-scale feature fusion network, and incorporating a hybrid attention mechanism, the recognition and counting accuracy of the algorithm for cells in complex microfluidic images is improved. The cell capture performance of the platform is systematically characterized to elucidate the influence of the Reynolds number (Re) and capture time on the capture yield, thereby establishing the optimal conditions for stable operation. On this basis, a stable linear model between cell-capture images of varying quality and cell concentrations within a specific range is established, and successfully used for Red Blood Cell (RBC) count prediction. Ultimately, this work provides a novel automated sensing platform that is reliable, efficient, and cost-effective for early screening and the monitoring of critical indicators in blood-related diseases.","[""Liu Yanwen"", ""Liu Zhaomiao"", ""Xuan Xinlong"", ""Zheng Kai"", ""Zhao Siyu"", ""Wang Xiang"", ""Pang Yan""]",Talanta,2026-10-08,10.1016/j.talanta.2026.130631,"[""Journal Article""]","[""Point-of-Care Systems"", ""Blood Cell Count"", ""Microfluidics"", ""Deep Learning""]", 42868808,https://pubmed.ncbi.nlm.nih.gov/42868808/,Multifunctional nanoflower-assisted catalytic hairpin assembly enables extracellular vesicles derived-microRNAs profiling for colorectal cancer diagnosis.,"The clinical translation of small extracellular vesicle-derived microRNAs (sEV-miRNAs) for early colorectal cancer (CRC) diagnosis is often limited by their low abundance and complexity of biological fluids. Herein, we developed an ultrasensitive electrochemical platform based on multifunctional DNA nanoflower-assisted catalytic hairpin assembly (NF-CHA) for sEV-miRNAs detection. The NF acts as a localized scaffold to accelerate CHA kinetics and load abundant signal reporter. As a result, the NF-CHA strategy enables the platform achieving a limit of detection of 23.65 aM. To improve diagnostic performance, we employed different sEV-miRNAs combination and machine learning (ML) algorithms, achieving an area under the curve (AUC) of 0.996 for distinguishing early-stage CRC from healthy donors in clinical samples. Therefore, the electrochemical platform holds a great promise as a robust tool for CRC early diagnosis.","[""Zhang Xiaohe"", ""Pan Jiezhen"", ""Lin Yuanhong"", ""Lin Zhixuan"", ""Zhu Yiping"", ""Yang Chao"", ""Zhang Ye"", ""Yan Lizhi"", ""Li Zihua"", ""Yan Xiaohui""]",Talanta,2026-10-07,10.1016/j.talanta.2026.130751,"[""Journal Article""]","[""Humans"", ""Colorectal Neoplasms"", ""MicroRNAs"", ""Extracellular Vesicles"", ""Electrochemical Techniques"", ""Biosensing Techniques"", ""DNA Nanostructures"", ""Catalysis"", ""Machine Learning"", ""Limit of Detection""]", 42863170,https://pubmed.ncbi.nlm.nih.gov/42863170/,Machine learning-assisted Ti3C2Tx-PANI-SnO2 heterojunctions for real-time monitoring of NH3 released from food spoilage.,"Ammonia (NH3) released during food spoilage is an important indicator of food freshness, yet its accurate monitoring remains challenging because of high humidity and interference from coexisting volatile amines. In this study, an intelligent sensing platform integrating a Ti3C2Tx-polyaniline (PANI)-SnO2 multidimensional heterojunction sensor with Transformer-based signal analysis was developed for reliable, real-time NH3 monitoring. The synergistic combination of Ti3C2Tx, PANI, and SnO2 enabled efficient signal transport, selective NH3 adsorption, enhanced interfacial charge transfer, and improved structural stability. The optimized Ti3C2Tx-PANI(50)-SnO2(30) sensor exhibited a response of 12.6 to 200 ppm NH3 over a wide detection range of 0.2-200 ppm. It maintained stable sensing performance at 20%-90% relative humidity and retained approximately 80% of its initial response after 60 days of storage, demonstrating good humidity tolerance and long-term stability. To mitigate interference from dimethylamine and trimethylamine generated during food spoilage, a Transformer model was employed to analyze the dynamic sensing signals. The model achieved a classification accuracy of 97.51% for mixed-gas samples and enabled reliable quantitative determination of NH3 under complex interference conditions. The sensor and embedded model were further integrated into a portable device for real-time monitoring of NH3 released during the spoilage of shrimp, fish, and pork. The measured NH3 concentrations deviated from the actual values by less than 10%. This multifunctional intelligent sensing platform provides a practical approach for on-site, intelligent food freshness assessment in complex high-humidity environments.","[""Su Chenglin"", ""Li Jiaxing"", ""Shen Lei"", ""Zheng Haotian"", ""Li Chen"", ""Liu Song"", ""Li Siqi"", ""Niu Na"", ""Chen Ligang""]",Talanta,2026-10-07,10.1016/j.talanta.2026.130752,"[""Journal Article""]","[""Ammonia"", ""Machine Learning"", ""Tin Compounds"", ""Aniline Compounds"", ""Titanium"", ""Animals"", ""Humidity""]", 42854366,https://pubmed.ncbi.nlm.nih.gov/42854366/,Online rapid detection of biomass in avilamycin fermentation process based on near-infrared spectroscopy technology.,"Biomass is one of the key parameters to be monitored during avilamycin fermentation. Currently, periodic sampling is commonly employed for biomass detection, which often entails significant manpower and cost expenditure. Therefore, it is crucial to develop a method capable of real-time, online monitoring of biomass changes during avilamycin fermentation. This study selected wet weight as the biomass measurement indicator during avilamycin fermentation, and employed online near-infrared spectroscopy (NIRS) technology to perform rapid, non-destructive monitoring of wet weight content throughout the fermentation process. Comparing the models' performance based on three algorithms, including partial least squares regression (PLSR), principal component regression (PCR) and support vector regression (SVR), it was confirmed that the PLSR model demonstrated optimal performance. After outlier removal, preprocessing method selection, and feature wavelengths screening, the results indicated that the dynamic biomass prediction model constructed using the SG+1st Der-i-PLSR method exhibited the optimal predictive performance. Within the validation set, the prediction set determination coefficient (R2p), root mean square error of prediction (RMSEP), residual predictive deviation (RPD), and standard error of prediction (SEP) were 0.9319, 21.19, 3.351 and 20.98, respectively. For wet weight content ranging from 109.5 to 436.5 g/L, the intuitive metric mean absolute error (MAE) was 16.30 g/L after external validation across five fermentation batches. The foregoing indicates that this model holds promising application prospects for online non-destructive biomass monitoring during avilamycin fermentation. It will lay the foundation for process optimization and control in the Industrial-scale production of avilamycin.","[""Zhou Bingyi"", ""Shen Yifan"", ""Chen Min"", ""Long Hongzhi"", ""Wang Zeyu"", ""Lin Wensen"", ""Jia Minghui"", ""Pan Xingru""]",Talanta,2026-09-21,10.1016/j.talanta.2026.130648,"[""Journal Article""]","[""Spectroscopy, Near-Infrared"", ""Fermentation"", ""Biomass"", ""Least-Squares Analysis"", ""Algorithms"", ""Support Vector Machine""]", 42837784,https://pubmed.ncbi.nlm.nih.gov/42837784/,Bimetallic FeZr-MOF with enhanced peroxidase activity via Fe-O-Zr bridging electronic synergistic effects: a three-channel sensor for organophosphorus pesticide recognition.,"Organophosphorus pesticides present a notable risk to public health, making it essential to develop swift and dependable analytical techniques for food safety, yet conventional single-channel colorimetric sensors face great challenges in discriminating multiple structurally similar pesticides with high accuracy. To address these issues, a Fe-Zr bimetallic metal-organic framework nanozyme with prominently enhanced peroxidase activity was synthesized via a one-step solvothermal method. The intermetallic electron transfer through Fe-O-Zr bridges accelerates the reversible Fe3+/Fe2+ redox cycle; with a Michaelis-Menten constant (Km) of 0.11 mM for H2O2 and a maximum reaction rate (Vmax) of 9.95 × 10-8 M s-1, its catalytic efficiency outperforms monometallic analogues and the majority of previously reported MOF-based nanozymes. On this basis, a three-channel sensor array integrating absorbance, smartphone-based RGB colorimetric and photothermal signals was constructed, coupled with principal component analysis (PCA) and hierarchical cluster analysis (HCA) for chemometric pattern recognition. The sensor works effectively over a wide concentration range (5-200 μM), with excellent interference resistance and satisfactory recoveries (94.2%-107.4%) in real samples. Most importantly, the Fe-O-Zr bridging synergistic strategy offers a new approach to enhance MOF peroxidase activity, while the multi-channel architecture mitigates the limitation of single-channel sensors in differentiating similar pesticides, supporting reliable on-site food safety and environmental monitoring.","[""Xu Ming"", ""Men Sihan"", ""Zheng Kai"", ""Tian Miaomiao""]",Talanta,2026-09-29,10.1016/j.talanta.2026.130710,"[""Journal Article""]","[""Pesticides"", ""Zirconium"", ""Colorimetry"", ""Metal-Organic Frameworks"", ""Organophosphorus Compounds"", ""Peroxidase"", ""Iron"", ""Hydrogen Peroxide"", ""Catalysis"", ""Oxidation-Reduction""]", 42822294,https://pubmed.ncbi.nlm.nih.gov/42822294/,Automated machine learning-Assisted interpretation of GC-MS urinary organic acid profiles by integrating explicit metabolic markers and chromatographic patterns for rare organic acidemias.,"Organic acidemias (OAs) are inherited metabolic disorders requiring accurate profiling of urinary organic acids for diagnosis. Although gas chromatography-mass spectrometry (GC-MS) remains a clinical standard, interpretation is challenging because urinary metabolic profiles are complex and heterogeneous, reflecting disease-related changes and biological variability arising from physiological, dietary, and genetic factors. Together with limited authentic derivatized reference materials, these factors make interpretation labor-intensive and dependent on expert judgment. We developed a feature-based machine-learning framework for OA interpretation under limited-sample conditions. GC-MS chromatographic patterns and metabolite response ratios from 727 urine samples (475 controls and 252 OA cases across 18 categories) were evaluated using convolutional neural networks (CNN), support vector machines (SVM), random forests (RF), and deep neural networks (DNN). All models achieved F1 scores >0.80 for OA discrimination, with DNN showing the highest performance (F1 = 0.84, AUC = 0.85). For the 13-class task, DNN achieved an accuracy of 0.89, F1 score of 0.90, and AUC of 0.98, followed by RF (F1 = 0.88, AUC = 0.97) and SVM (F1 = 0.85, AUC = 0.95). Notably, our framework used only 34 pathway-informed features from 727 samples, whereas previous GC-MS machine-learning studies used 132 metabolites from 531 samples and 135 metabolites from 355,197 samples, reporting accuracies of >0.981 and 0.969, respectively, under different classification settings. These findings demonstrate that a compact, biochemically interpretable feature set can capture disease-associated patterns under data-limited conditions and facilitate consistent GC-MS interpretation, providing complementary support for expert OA diagnosis.","[""Ma Chin-Hsiang"", ""Jair Yung-Cheng"", ""Chen Ju-Yu"", ""Chen Hui-An"", ""Hsu Rai-Hseng"", ""Weng Te-I"", ""Chien Yin-Hsiu"", ""Hwu Wuh-Liang"", ""Lai Feipei"", ""Lee Ni-Chung"", ""Chen Pai-Shan""]",Talanta,2026-09-30,10.1016/j.talanta.2026.130711,"[""Journal Article""]","[""Gas Chromatography-Mass Spectrometry"", ""Humans"", ""Machine Learning"", ""Support Vector Machine"", ""Random Forest"", ""Biomarkers"", ""Convolutional Neural Networks"", ""Neural Networks, Computer"", ""Classification Algorithms""]", 42815256,https://pubmed.ncbi.nlm.nih.gov/42815256/,Prototypical transfer learning framework for small-sample 1H NMR metabolomics: Accurate and interpretable classification of short stature.,"Short stature (SS) is a common pediatric endocrine disorder, primarily including growth hormone deficiency (GHD) and idiopathic SS (ISS). Nuclear magnetic resonance (NMR)-based metabolomics can reveal metabolic signatures associated with SS. However, the limited availability of clinical data results in small sample sizes and high-dimensional features, which constrain the stability and generalizability of traditional methods. In this study, a prototypical transfer learning approach for small-sample metabolomics data is proposed. By integrating a shared feature encoder, cross-domain feature alignment, and a prototype learning strategy, knowledge from the source domain (child malnutrition data) is effectively transferred to the target domain (SS data), mitigating domain shift and enhancing classification performance. Experimental results show that the proposed method achieves classification accuracies of 96.76% and 95.12% on the preadolescent and adolescent SS datasets, respectively, substantially outperforming traditional machine learning and deep learning models trained solely on the target domain. Interpretability analysis using Integrated Gradients indicates that the model primarily focuses on chemical shift regions corresponding to disease-related metabolites, confirming the biological relevance of the discriminative features. This study provides a novel and interpretable framework for small-sample metabolomics analysis and clinical diagnosis of SS.","[""Ye Zihang"", ""Ji Zhicheng"", ""Xia Feng"", ""Zhu Hongwei"", ""Feng Jianghua"", ""Shen Guiping""]",Talanta,2026-09-29,10.1016/j.talanta.2026.130677,"[""Journal Article""]","[""Humans"", ""Metabolomics"", ""Child"", ""Proton Magnetic Resonance Spectroscopy"", ""Growth Disorders"", ""Magnetic Resonance Spectroscopy""]", 42804948,https://pubmed.ncbi.nlm.nih.gov/42804948/,AI-assisted smartphone-based prompt fluorescence imaging for adulteration monitoring of extra virgin olive oil.,"Detection of adulteration in extra-virgin olive oil (EVOO) and tracing possible sources remain critical challenges for food authenticity, consumer safety, and regulatory enforcement, particularly in resource-limited settings where the application of high-accuracy analytical tools is impractical. In this work, a practical integration of a machine learning system with a compact, ultra-low-cost smartphone-based prompt fluorescence imaging platform is presented for on-site detection of EVOO adulteration, where prompt fluorescence refers to fluorescence recorded during excitation without time-delayed detection. The proposed platform employs a λ= 405 nm laser excitation source and orthogonal fluorescence imaging geometry, customized with a smartphone CMOS camera. The smartphone runs as a two-stage analytical framework, comprising adulterant classification followed by concentration estimation. The performance of machine-learning approaches (LightGBM, SVM, KNN, RF, Logistic Regression, Naive Bayes, Gradient Boosting, Decision Tree, Lasso Regression, SVR) has been evaluated using EVOO adulterated with common sources of adulteration such as refined olive oil, soybean oil, and palm oil over a wide concentration range. Machine learning models attain strong discrimination ability and high predictive accuracy by utilizing non-linear relationships within the fluorescence-derived RGB features. Independent validation using real samples confirms reliable adulterant identification and accurate EVOO purity estimation, with an average error of 3.1%. The present framework has been developed and validated using binary mixtures containing EVOO and a single adulterant, with the minimum operational detection threshold being 2% adulterant concentration. The proposed approach overcomes key limitations of conventional methods by enabling simultaneous detection of adulterant type and quantification without prior knowledge of the adulterant type, and therefore allowing users to predict the possible source of adulteration.","[""Talukder Arnab"", ""Al Rayhan Md Sadik"", ""Ifty Mohiminur Rahman"", ""Joarder Md Tanjim"", ""Hossain Md Arafat"", ""Islam Md Rafiqul"", ""Biswas Protik Chandra""]",Talanta,2026-09-25,10.1016/j.talanta.2026.130673,"[""Journal Article""]","[""Olive Oil"", ""Food Contamination"", ""Smartphone"", ""Machine Learning"", ""Optical Imaging"", ""Fluorescence"", ""Food Analysis""]", 42800159,https://pubmed.ncbi.nlm.nih.gov/42800159/,A review on lateral flow immunoassay technology for mycotoxin detection: From single-signal enhancement to multi-signal synergy.,"Mycotoxins are toxic secondary metabolites produced by fungi that frequently contaminate grains, nuts, and other foodstuffs. Their widespread occurrence causes significant economic losses and poses a serious threat to human and animal health. Traditional analytical methods relying on complex instruments often fail to meet the requirements for rapid on-site screening and remain limited by cost, operational complexity, and restricted on-site deployment capabilities. Lateral flow immunoassays (LFIAs) are important tool for the rapid detection of mycotoxins due to their simplicity, portability, and low cost. This review provides a critical overview of the latest development in signal readout architecture for LFIA platforms, including single-signal platforms based on colorimetry, fluorescence, chemiluminescence, and surface enhanced Raman scattering (SERS); dual-mode system integrating colorimetric-fluorescence or colorimetric-photothermal readings; and multi-mode platforms that combines three or more orthogonal signals. It also discusses the mechanism of functional nanomaterials in signal amplification, with particular emphasis on gold and fluorescent nanomaterials, nanozymes, metal organic frameworks, magnetic nanoparticles, and layered assembled plasmonic probes. The integration of portable readers, smartphone imaging, and machine learning algorithms, and high-throughput multiplexing formats is also summarized. In summary, future priorities have been identified, including multifunctional nanoprobes, intelligent on-site deployment devices, and AI assisted data interpretation. This review provides a systematic framework for the rational design of highly sensitive, reliable, and field-deployable LFIA technologies for mycotoxin monitoring.","[""Li Menghan"", ""Zhang Chenjin"", ""Huang Dong"", ""Wei Yu"", ""Tao Jing"", ""Wang Yudan"", ""Wang Meng""]",Talanta,2026-09-24,10.1016/j.talanta.2026.130660,"[""Journal Article"", ""Review""]","[""Immunoassay"", ""Mycotoxins"", ""Colorimetry"", ""Humans"", ""Spectrum Analysis, Raman"", ""Food Contamination""]", 42800152,https://pubmed.ncbi.nlm.nih.gov/42800152/,Label-free and intelligent fluorometric array for simultaneously sensing of multiplexed metal ions based on DNA conformational diversity.,"Multiplex detection of metal ions is often challenged by their similar charge states and ionic radii, which interfere with for simultaneous analysis and cause the low selectivity of conventional lock-and-key recognition elements. Herein, a label-free and intelligent fluorometric array has been developed for simultaneously probing multiplexed metal ions by utilizing the conformational diversity of DNA structures, including G-triplex, G-quadruplex, and dimeric G-quadruplex DNA. The sensing mechanism relies on thioflavin T intercalation, which converts metal-binding-induced DNA conformational changes into quantifiable fluorescence signals. The resulting fluorescence fingerprints were first visualized using principal component analysis and subsequently classified using supervised machine-learning algorithms. A comparative evaluation of multiple models reveals that the support vector machine achieves the highest classification performance for selective identification of multiple metal ions. The developed fluorometric array can accurately distinguish the samples containing various metal species (Fe3+, Co2+, Hg2+, Ni2+, Zr4+, Ag+, and Zn2+) over a wide range of concentrations (10-200 μM), with a 100% accuracy in blind validation tests. Subsequently, the feasibility of the array for detecting multiple metal ions in mixtures, as well as in lake water and tap water, is also demonstrated. As compared with most current analysis technologies, this method offers the desirable advantage of simultaneous discrimination of multiple metal ions including the heavy and non-heavy ones. Therefore, this study establishes a powerful, efficient, and cost-effective platform for the simultaneous and intelligent detection of multiple metal ions, showing strong potential as a rapid on-site screening tool for preliminary risk assessment of metal contamination.","[""Shi Lu"", ""Jin Jiayi"", ""Liu Hanyao"", ""Wu Na"", ""Xu Chuyun"", ""Bu Jianghao"", ""Zhao Xiaoting"", ""Li Shuai"", ""Qi Yanxia"", ""Li Shanshan"", ""Wang Hua""]",Talanta,2026-09-22,10.1016/j.talanta.2026.130656,"[""Journal Article""]","[""Fluorometry"", ""DNA"", ""Metals"", ""G-Quadruplexes"", ""Nucleic Acid Conformation"", ""Benzothiazoles"", ""Ions"", ""Support Vector Machine"", ""Principal Component Analysis""]", 42786613,https://pubmed.ncbi.nlm.nih.gov/42786613/,Neuroimaging Biomarkers for Brain Disorder Diagnosis Using Advanced Learning Techniques.,"Mental illness is becoming increasingly common in everyday life, affecting a growing number of people. Functional Magnetic Resonance Imaging (fMRI) is an effective technique for detecting mental disorders. Resting-state fMRI (rs-fMRI) analyses spontaneous low-frequency oscillations in the Blood Oxygen Level Dependent signal to examine the functional architecture of the brain. fMRI and rs-fMRI data processing contribute to mapping neural activity in specific brain regions, enabling accurate localization of cognitive functions and improving our understanding of Functional Connectivity and interactions among brain regions. This paper focuses on the use of Machine Learning and Deep Learning techniques for fMRI analysis to identify brain features that may be associated with Bipolar Disorder (BD) and Alzheimer's Disease (AD). The first step in the proposed approach is the preprocessing of fMRI and rs-fMRI data using two distinct methods: the first combines Statistical Parametric Mapping and the CONN toolbox for improved results, and the second focuses on the MELODIC software. The proposed approach begins with fMRI and rs-fMRI preprocessing using two pipelines: one integrating Statistical Parametric Mapping (SPM) with the CONN toolbox and another based on the MELODIC framework. The aim is to evaluate how preprocessing choices affect functional connectivity representations and classification performance when combined with classical Machine Learning (ML) methods, namely Support Vector Machines and Random Forests, and the Deep Learning (DL) model AlexNet. Given the altered brain connectivity patterns observed in Bipolar Disorder and Alzheimer's Disease, these methods are employed to assess preprocessing-learning interactions across both disorders. Experimental results emphasize that Connbox-SPM preprocessing combined with AlexNet yields the best results for both mental disorders, with Bipolar Disorder reporting 78.2% accuracy and AD achieving 82.4% accuracy.","[""Ion Iulia-Andreea"", ""Chira Camelia""]",International journal of neural systems,2026-09-25,10.1142/S0129065727500158,"[""Journal Article""]","[""Humans"", ""Bipolar Disorder"", ""Magnetic Resonance Imaging"", ""Alzheimer Disease"", ""Machine Learning"", ""Deep Learning"", ""Brain"", ""Random Forest"", ""Support Vector Machine"", ""Neuroimaging"", ""Convolutional Neural Networks"", ""Classification Algorithms"", ""Predictive Learning Models""]", 42671319,https://pubmed.ncbi.nlm.nih.gov/42671319/,Semantic-Aware Semi-Supervised Systematic Segmentation for Brain Magnetic Resonance Image.,"Accurate segmentation of brain MRI images is essential for early diagnosis and prevention of brain tumors and strokes. However, it is challenging due to data scarcity, domain shift, complexity, and blurred edges. Recent methods use mean-teacher model learning for the semantic details of an individual image based on the efficiency of teacher architecture. However, to enhance the segmentation in brain MRI, the paper introduces SemS4, a semi-supervised method on the basis of cross-image architecture. SemS4 uses a single architecture integrating an Edge Prototype Attention (EPA) and a foreground prototype attention (FPA). More specifically, EPA uses edge prototype integrated with adaptive edge container to improve the edge features for which global edge features are stored to produce edge prototype. Moreover, FPA stabilizes the segmentation by transferring complementary foreground information. Furthermore, the proposed method uses Pixel Affinity Loss (PAL) for missing contextual correlation during supervision to enhance segmentation performance on edges. The performance of the proposed approach in experimental results for two brain MRI benchmarks, LGG and BRISC, validates the superiority over existing methods under different partitioning settings.","[""Madni Hussain Ahmad"", ""Zottin Silvia"", ""Nardin Axel De"", ""Umer Rao Muhammad"", ""Foresti Gian Luca""]",International journal of neural systems,2026-08-31,10.1142/S0129065727500183,"[""Journal Article""]","[""Magnetic Resonance Imaging"", ""Humans"", ""Brain"", ""Semantics"", ""Brain Neoplasms"", ""Image Processing, Computer-Assisted"", ""Supervised Machine Learning""]", 42602370,https://pubmed.ncbi.nlm.nih.gov/42602370/,EINN: An enzyme-informed neural network guided by an enzyme-constrained genome-scale metabolic model.,"Predicting cellular metabolism from molecular data is a central challenge in systems biology, with direct implications for engineering microbial cell factories. Genome-scale metabolic models (GEMs) provide a mechanistic foundation for such predictions, but their accuracy is limited by an inability to account for the finite catalytic capacity of the proteome. Enzyme-constrained GEMs (ecGEMs) address this by incorporating enzyme turnover numbers and proteome allocation constraints. Hybrid neural-mechanistic models like Artificial Metabolic Network (AMN) and Metabolic-Informed Neural Network (MINN) have sought to combine the flexibility of machine learning with the structure of GEMs, yet they rely on conventional, unconstrained metabolic networks, allowing flux predictions to violate enzyme capacity constraints and pushing models toward biologically unrealistic solution spaces. Here, we introduce the Enzyme-Informed Neural Network (EINN), a conceptual framework which relies on ecGEMs to constrain the neural network. We implement two variants of EINN, ecAMN and ecMINN, using Escherichia coli (E. coli) ecGEMs built with the GECKO 3.0 toolbox, and systematically evaluate their performance. Compared to their unconstrained counterparts, ecAMN significantly improves training stability and predictive accuracy by eliminating convergence to poor local minima, while ecMINN reduces overfitting and achieves lower error rates through mechanistic integration of proteomic data as flux bounds. Together, these results show that constraining the neural network's solution space with enzymatic capacity limits is a more effective hybrid modeling foundation than relying on conventional GEMs alone.","[""Steven Ray"", ""Kimata Kai"", ""Chegodaev Denis"", ""Handayani Lilies"", ""Satou Kenji""]",Synthetic and systems biotechnology,2026-08-06,10.1016/j.synbio.2026.07.009,"[""Journal Article""]",,PMC13474246 42962522,https://pubmed.ncbi.nlm.nih.gov/42962522/,Chemically informed spectro-structural feature augmentation for interpretable NIR chemometric classification of fuel grades.,"Rapid and interpretable near-infrared discrimination of fuel grades remains challenging because adjacent fuel grades usually exhibit highly overlapped C-H overtone and combination bands, while the structural origin of model decisions is often insufficiently described. To address this problem, this study proposes a knowledge-guided spectro-structural feature augmentation framework, termed SSFA, for interpretable NIR-based fuel classification. Unlike conventional variable selection or simple feature-weighting methods, SSFA uses changes in the carbon-chain length of n-alkanes as chemical prior knowledge, converts the ""carbon chain length-NIR response"" relationship into structure-augmented features with directional constraints, and embeds structure-sensitive bands into the model input space as learnable descriptors. These chemically meaningful descriptors are fused with full-spectrum NIR variables, enabling classifiers to simultaneously capture data-driven spectral patterns and structural discriminative cues. A total of 292 fuel samples were collected, covering jet fuel, gasoline, and diesel from 26 representative refineries. The effectiveness of SSFA was systematically evaluated under different spectral preprocessing methods, feature selection strategies, and machine-learning classification models. SSFA improved classification performance across multiple linear, nonlinear, and tree-based models. The optimal model, based on SG second-derivative preprocessing and BP-ANN, achieved a fuel-grade recognition accuracy of 95.6% and an average positive predictive value of 96.30%. In addition, all 12 real fuel samples were correctly identified, providing preliminary evidence for practical applicability. The results indicate that SSFA is an interpretable feature augmentation framework for NIR-based fuel quality screening and may provide a methodological reference for spectral classification of other complex chemical mixtures.","[""Zhang Xin"", ""Li Ke"", ""Liu Fan"", ""Yan Yong"", ""Zhang Zhengdong"", ""Cheng Yingying""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-27,10.1016/j.saa.2026.129055,"[""Journal Article""]",, 42925695,https://pubmed.ncbi.nlm.nih.gov/42925695/,Fabrication of carbon dots derived from Alangium salviifolium for the sequential sensing Ag+ and S2- and further monitoring of H2S in spoiled food.,"Given the significant threat posed by heavy metals to ecological system and human health, the development of precise, portable, and on-site analytical technologies is urgently required. Herein, multifunctional carbon dots (A-CDs) derived from the traditional chinese medicine (TCM) were synthesized and exhibited optimal emission at 604 nm excited at 495 nm, with quantum yield (QY) of 18.25%. The A-CDs achieved the cascade determination of Ag+ and S2- via a subsequent ""on-off-on"" process with LODs of 63.7 nM and 0.98 μM, respectively. Notably, A-CDs portable platform was constructed and enabled on-site quantitative analysis with LODs of 0.83 μM and 1.19 μM, respectively. Leveraging this unique sensing performance, the A-CDs test strips were developed for visual screening of H2S during food spoilage at varying storage temperatures, allowing qualitative food freshness discrimination via naked-eye observation. The precise quantitative detection in real water and foodstuff was applied, and further verified by ICP-MS, confirmed the excellent accuracy and environmental tolerance. This work presented an intelligent strategy for designing cost-effective, high-performance multivariate probes, holding significant promise for advancing food safety and environmental monitoring.","[""Qi Xiukun"", ""Yang Huiyi"", ""Xu Dehong"", ""Cui Tingting"", ""Dai Qijun"", ""Chen Guangying"", ""Su Zhongmin"", ""Chai Fang""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-18,10.1016/j.saa.2026.128997,"[""Journal Article""]","[""Carbon Quantum Dots"", ""Silver"", ""Hydrogen Sulfide"", ""Limit of Detection"", ""Food Analysis"", ""Spectrometry, Fluorescence"", ""Carbon""]", 42925688,https://pubmed.ncbi.nlm.nih.gov/42925688/,Lightweight interpretable support vector machine SERS (LiSVM-SERS) learning framework for serological diagnosis of psoriatic arthritis and biomarker identification.,"Psoriatic arthritis (PsA), a chronic, autoimmune form of arthritis that causes joint inflammation and occurs with the skin condition psoriasis, is difficult to clinically diagnose at an early stage due to insignificant symptoms and lack of biomarkers. Surface-enhanced Raman spectroscopy (SERS) is a powerful tool to identify specific biomolecules for liquid biopsy. Here, we establish a lightweight interpretable support vector machine SERS (LiSVM-SERS) learning framework for serological diagnosis of PsA and identification of potential biomarkers. The fs-laser-nanofabricated Au@mica SERS substrates with a relative standard deviation of 8.80% in intensity homogeneity are employed to obtain reliable serum Raman spectra. The SVM algorithm is performed to train the machine-learning model for PsA classification using only 26 samples (PsA = 13 and healthy controls = 13). The hyperplane coefficient vectors from the trained SVM model are then employed as the filtered weights of Raman shifts, followed by logistic-probability mapping and Welch's t-test to filter highly-significant spectral features, identifying 432 most-significant Raman shifts for 7 potential biomarkers. The trained LiSVM-SERS model demonstrates the outstanding sensitivity and specificity of 98.0% and 96.0%, respectively, validated by an internal independent blind-test cohort (PsA = 30, healthy controls = 30) and a cross disease-control cohort (systemic lupus erythematosus = 15, rheumatoid arthritis = 15, and osteoarthritis = 15). The present work not only provides a simple and lightweight non-invasive technique to diagnose PsA, but also paves a new way to identify the potential biomarkers from the label-free serum SERS spectra for autoimmune disorders.","[""Zhang Huijuan"", ""Lin Wenhao"", ""Pan Hang"", ""Cao Lulu"", ""Wang Yan"", ""Sun Haoran"", ""Zhao Yan"", ""Jia Yuan"", ""Yan Yinzhou""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-20,10.1016/j.saa.2026.129004,"[""Journal Article""]","[""Spectrum Analysis, Raman"", ""Humans"", ""Arthritis, Psoriatic"", ""Biomarkers"", ""Support Vector Machine"", ""Female"", ""Male"", ""Sensitivity and Specificity"", ""Serologic Tests""]", 42914810,https://pubmed.ncbi.nlm.nih.gov/42914810/,Intelligent identification technology of zidovudine and its isomers based on active low-frequency terahertz excitation.,"Zidovudine, adenosine, and vidarabine are three isomeric compounds with highly similar physicochemical properties, making them prone to confusion during pharmaceutical manufacturing, storage, and transportation. Therefore, the development of identification methods with high specificity and sensitivity for such isomeric compounds is of great significance for ensuring drug quality and clinical safety. However, conventional detection methods suffer from limitations such as high cost, long processing time, or insufficient resolution, rendering drug identification a challenging issue. This study introduces a non-destructive and intelligent detection method by integrating terahertz time-domain spectroscopy (THz-TDS) with machine learning techniques. Transmission spectra of zidovudine, adenosine, and vidarabine were collected using a THz-TDS system and converted into absorbance spectra. Key features were extracted via t-distributed stochastic neighbor embedding (t-SNE), isometric mapping (Isomap), and principal component analysis (PCA). These were used to train support vector machine (SVM) and k-nearest neighbors (KNN) classifiers. Among the obtained models, the PCA-KNN combination achieved the most outstanding performance, with a classification accuracy of 100%, realizing precise discrimination of the three isomers. This method not only meets the quality control requirements of pharmaceutical standards and industrial production but also serves as a valuable reference for THz spectral detection of related drugs. It demonstrates promising potential for applications in precise drug identification and recognition.","[""Lu Chaoyue"", ""Pang Senhao"", ""Li Yanxin"", ""Su Jingkai"", ""Tan Kunwu"", ""Song Zhongzhou"", ""Tu Shan"", ""Tang Yuan"", ""Zheng Junjian"", ""Wang Xueye""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-20,10.1016/j.saa.2026.128916,"[""Journal Article""]","[""Zidovudine"", ""Terahertz Spectroscopy"", ""Support Vector Machine"", ""Isomerism"", ""Principal Component Analysis"", ""Vidarabine"", ""Adenosine"", ""Machine Learning""]", 42913026,https://pubmed.ncbi.nlm.nih.gov/42913026/,"Ortho-, meta- and para-substituted fluorescent probes based on BODIPY for detecting esterase with application in machine learning and bioimaging.","Esterase plays a significant role in industrial production and life activities. The method of fluorescent probe is employed for detecting esterase due to their excellent performance. However, the restriction of recognition groups to specific positions severely limits the molecular design and performance optimization of fluorescent probe. Thus, three fluorescent probes (OBOD6, MBOD6 and PBOD6) based on BODIPY with recognition groups at ortho-, meta- and para- positions were designed and synthesized to systematically investigate the positional effects on performance of probes. All probes initially showed fluorescence quenching, followed by characteristic enhancement of fluorescent intensity at 587 nm (OBOD6), 577 nm (MBOD6) and 598 nm (PBOD6) upon esterase-catalyzed hydrolysis. Remarkably, OBOD6 exhibited the fastest reaction kinetics, nearly achieving conversion within 40 min at esterase of 0.6 U/mL. When incubated with esterase of 0.3 U/mL, MBOD6 (10 μM) nearly reached the reaction endpoint within 80 min, exhibiting the highest fluorescent enhancement and optimal detection sensitivity. MBOD6 could quantitatively detect esterase within a concentration range from 0.01 to 0.07 U/mL, with a detection limit of 2.23 × 10-4 U/mL. Besides, machine learning was additionally applied to the quantitative detection of esterase using MBOD6. The model explained approximately 93.19% of the variation in esterase concentration, indicating good overall fitting capability. Finally, leveraging its outstanding performance, MBOD6 was successfully applied to imaging of esterase in both cells and zebrafish.","[""Liu Cong"", ""Liyang Muchi"", ""Bi Yunshuai"", ""Wang Yipei"", ""Zhang Tianyu"", ""Sun Bowen"", ""Guo Huichen"", ""Li Xinran"", ""Qin Wenwu""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-16,10.1016/j.saa.2026.128995,"[""Journal Article""]","[""Boron Compounds"", ""Esterases"", ""Fluorescent Dyes"", ""Animals"", ""Zebrafish"", ""Machine Learning"", ""Spectrometry, Fluorescence"", ""Kinetics"", ""Fluorescent Chemosensor Compounds"", ""Hydrolysis""]", 42893674,https://pubmed.ncbi.nlm.nih.gov/42893674/,Leveraging infrared spectral signatures supported by machine learning to evaluate the patency of coronary artery bypass graft conduits: A preliminary study.,"Coronary artery bypass grafting (CABG) is a critical intervention for obstructive coronary artery disease, with the graft patency significantly impacting long-term outcomes. This study evaluates the potential of infrared (IR) spectroscopic imaging combined with machine learning to assess biochemical markers for determining the patency in internal mammary artery (IMA), radial artery (RA), and saphenous vein (SV) conduits. Tissue samples of the grafts from CABG patients were analysed using IR spectroscopic imaging. Hyperspectral images were pre-processed and subjected to unsupervised hierarchical cluster analysis (UHCA) to identify regions with high triglyceride content. Spectral signatures of triglycerides, proteins, nucleic acids, and carbohydrates were extracted, and ratiometric analysis revealed significant variations in glycosylation and protein-to-lipid ratios among graft types (p < 0.005). The IMA exhibited lower glycosylation and protein-to-lipid ratios compared to the RA and SV. Principal component analysis followed by linear discriminant analysis (PCA-LDA) achieved clear differentiation of graft types, with second-derivative spectra enhancing class separation. These findings highlight IR spectroscopy's potential to provide detailed biochemical insights for pre-operative graft selection, potentially improving CABG outcomes by identifying conduits with optimal patency characteristics. This preliminary study evaluates the feasibility of IR spectroscopy for assessing the quality of autologous bypass grafts prior to CABG. Further studies involving larger patient cohorts and evaluation of potential gender-related differences would help strengthen and validate these findings.","[""Nazeer Shaiju S"", ""Rekha C R"", ""Smith Julian A"", ""McNaughton Don"", ""Jayasree Ramapurath S""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-13,10.1016/j.saa.2026.128962,"[""Journal Article""]","[""Coronary Artery Bypass"", ""Humans"", ""Machine Learning"", ""Spectrophotometry, Infrared"", ""Vascular Patency"", ""Principal Component Analysis"", ""Discriminant Analysis"", ""Cluster Analysis"", ""Clustering Algorithms""]", 42892243,https://pubmed.ncbi.nlm.nih.gov/42892243/,Raman spectroscopy coupled with interpretable machine learning for early detection of citrus Huanglongbing.,"Citrus Huanglongbing (HLB), caused by Candidatus Liberibacter asiaticus, threatens global citrus production, while early diagnosis is hindered by a prolonged asymptomatic period and laboratory-intensive methods. This study establishes a non-destructive framework integrating portable Raman spectroscopy, reference compound validation, optimized machine learning, and model interpretability for early HLB detection. A total of 1025 Thai pomelo leaves yielding 3075 spectra were classified into four categories: healthy, early HLB, symptomatic HLB, and nutrient deficiency. Class-specific spectral variations occurred in carotenoid, chlorophyll a, carbohydrate, and polyphenol regions. Measurements of pure reference standards supported key molecular assignments beyond literature-based interpretation. UMAP followed by K-means clustering achieved a purity of 91.72% ± 0.96%, indicating spectral separability among the four classes. For supervised classification, 72 preprocessing workflows were evaluated across six classifiers, followed by hyperparameter optimization using four search strategies. The optimized support vector machine (SVM) achieved the highest accuracy of 95.93% ± 1.42% and an F1-score of 0.9597 ± 0.0146. SHAP analysis linked classification decisions to these molecular assignments, identifying pigment-associated features as major contributors to overall classification and a cellulose-associated feature at 1096.8 cm-1 as the leading contributor to early HLB identification. By connecting molecular assignments, classification performance, and model interpretation, this framework provides a biologically grounded approach to early HLB detection with potential for field deployment.","[""Hu Xin"", ""Zhang Jixing"", ""Chu Yuexiang"", ""Gao Zhimin"", ""Xiong Pei"", ""Li Shanjun"", ""Yi Weisong""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-12,10.1016/j.saa.2026.128959,"[""Journal Article""]","[""Spectrum Analysis, Raman"", ""Citrus"", ""Plant Diseases"", ""Machine Learning"", ""Plant Leaves"", ""Support Vector Machine"", ""Classification Algorithms""]", 42880085,https://pubmed.ncbi.nlm.nih.gov/42880085/,Multiscale transformer model based on Raman spectroscopy detection in diagnosis of colorectal cancer.,"INTRODUCTION: There is currently a lack of low-cost, nonintrusive methods for colorectal adenocarcinoma (CRA) detection. Moreover, the application of Raman spectroscopy (RS)-based CRA detection is limited due to its high noise levels and low characterization abilities. OBJECTIVES: Propose a method for the early screening of CRA based on RS and deep learning techniques. METHODS: An MS-Former (multiscale transformer) method was constructed by combining multiscale convolution and self-attention mechanisms to adapt spectral scenarios based on RS and surface-enhanced Raman spectra (SERS) at the cellular and patient serum levels, respectively. Furthermore, Raman spectral critical frequency test (classifying the spectrum extracted from each threshold by linear discriminant analysis (LDA) using the classical machine learning method) was performed and the concerning Raman shifts were screened in the identification process of different samples. RESULTS: At the cellular level, the proposed model can identify normal cells and cancer cells at an accuracy of 99.12%. At the serum level, for the first time, this model achieved AUC values of 98.45% and 99.40% in identification of adenoma and adenocarcinoma, respectively. When classifying patients into groups of healthy, adenoma, and adenocarcinoma, the single Raman detection and individual level accuracy were 94.78%, 100%, respectively. Matching metabolites were obtained through matching the screened Raman shifts, which greatly improved the interpretability and biological significance of the results. CONCLUSION: The proposed model can conveniently and rapidly screens patients with colorectal adenomas and adenocarcinomas, requiring no pretreatment and exhibiting extremely high accuracy and low cost.","[""Xu Jiasheng"", ""Zheng Xingyan"", ""Hu Yeting"", ""Li Chao"", ""Chen Xueyan"", ""Dai Siqi"", ""Kong Xiangxing"", ""Ding Lei"", ""Wang Ben"", ""Ding Kefeng""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-23,10.1016/j.saa.2026.128756,"[""Journal Article""]","[""Spectrum Analysis, Raman"", ""Humans"", ""Colorectal Neoplasms"", ""Discriminant Analysis"", ""Adenocarcinoma""]", 42874475,https://pubmed.ncbi.nlm.nih.gov/42874475/,Fe3+-mediated fluorescence switching of biomass-derived carbon dots integrated with PVA microneedles for smartphone-assisted determination of ascorbic acid in fruits and vegetables.,"Rapid and accurate determination of ascorbic acid (AA) is important for evaluating the nutritional quality of fresh produce. In this study, biomass-derived carbon dots (CDs) were synthesized from pomelo peel in a green deep eutectic solvent, and Fe3+ was subsequently coordinated with the CDs to construct an off-on fluorescent probe based on Fe3+-induced quenching and AA-triggered fluorescence recovery. The probe was incorporated into a polyvinyl alcohol (PVA) microneedle (MN) patch to enable localized sampling and in situ sensing in fruits and vegetables. Fluorescence images were acquired using a smartphone, and multidimensional RGB-derived features were analyzed using LightGBM for quantitative prediction. The platform exhibited a linear response to AA over 0-10 mM (R2 = 0.9979) with a limit of detection of 0.2 μM, while the direct sampling, fluorescence sensing, and smartphone readout could be completed within 10 min. In real produce samples, the microneedle measurements showed good agreement with the reference method, supporting the potential of this portable platform for on-site AA quantification.","[""Hu Jing"", ""Hao Tong"", ""Sun Qianran"", ""Chen Mingrui"", ""Liu Yaowen""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-09,10.1016/j.saa.2026.128950,"[""Journal Article""]","[""Ascorbic Acid"", ""Carbon Quantum Dots"", ""Fruit"", ""Polyvinyl Alcohol"", ""Vegetables"", ""Needles"", ""Spectrometry, Fluorescence"", ""Smartphone"", ""Biomass"", ""Limit of Detection"", ""Fluorescence"", ""Carbon"", ""Fluorescent Dyes"", ""Fluorescent Chemosensor Compounds""]", 42874472,https://pubmed.ncbi.nlm.nih.gov/42874472/,Machine-learning-assisted ultraviolet-visible spectroscopy for low-level multi-component adulteration identification in Coptis chinensis.,"Low-level adulteration of botanical ingredients is difficult to detect because weak adulterant signals are often masked by the dominant matrix spectrum. Here, Ultraviolet-Visible (UV-Vis) spectroscopy combined with machine learning was investigated for identification of low-level multi-component adulteration in Coptis chinensis. Scutellaria baicalensis, Rheum palmatum, and Curcuma longa were used as adulterants to construct 29 classes, including pure, single-, binary- and ternary-adulteration samples, at a fixed total adulterant content of 3 wt%. Twenty independently prepared samples were included in each class, resulting in a total of 580 independent samples for model development and evaluation. The spectra showed strong overlap, nearly unchanged peak positions, and subtle intensity variations, making visual inspection and single-wavelength analysis insufficient for detailed discrimination. Multiple spectral preprocessing methods and machine-learning models were systematically compared using stratified five-fold cross-validation at the independent sample level. Following this systematic comparison, moving-average preprocessing combined with either random forest or extreme gradient boosting achieved the highest overall classification accuracy of 95.9%. Spectral-window analysis further showed that broader wavelength regions provided higher accuracy than narrow peak-centered windows, indicating that the discriminative information is distributed across the spectral profile rather than confined to isolated absorption peaks. Even though the study was limited to three model adulterants under the present controlled laboratory conditions, these results demonstrate the feasibility of extracting weak composition-dependent information from highly overlapping UV-Vis spectra and highlight the need for further validation using more diverse adulterants and sample sources.","[""Zhou Chao"", ""Jin Yaxuan"", ""Yang Ning"", ""Jiang Xianguo"", ""Zhang Shaohua"", ""Hao Lei"", ""Feng Dan"", ""Lu Chao"", ""Xu Jia"", ""Huang Wendeng""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-08,10.1016/j.saa.2026.128951,"[""Journal Article""]","[""Spectrophotometry, Ultraviolet"", ""Machine Learning"", ""Coptis"", ""Drug Contamination""]", 42868758,https://pubmed.ncbi.nlm.nih.gov/42868758/,Construction of a novel index driven by time-series hyperspectral features and the PROSAIL model for early warning of walnut leaf scorch.,"Walnut leaf scorch syndrome is a critical factor hindering the high-quality development of the walnut industry. Its incubation period is characterized by a lack of identifiable external morphological features, resulting in significant delays in early warning through traditional monitoring methods. This study seeks to establish a proactive early identification system for scorch leaf syndrome by integrating temporal spectral features with physical mechanism models. The goal is to provide theoretical support and technical assurance for precise forecasting, as well as proactive prevention and control measures in orchards. The study focused on the 'Wen 185' walnut plot in Aksu Prefecture and employed hyperspectral dynamic monitoring throughout the entire growing season. By applying the PROSAIL radiative transport model to derive physiological indicators of leaves, a novel physically driven index was developed. A comprehensive comparison was conducted on the performance of genetic algorithm, simulated annealing algorithm, and random forest in managing time-series redundant bands and identifying key features. Ultimately, models utilizing random forest, support vector machine, and artificial neural network were established for the early detection of focal leaf syndrome. The findings indicate that, despite the absence of visible lesions on walnut leaves during the latent period of scorch leaf syndrome, the hyperspectral curve has already exhibited a distinct ""blue shift"" trend at the red edge position. At the feature selection level, genetic algorithms, due to their robust global search capabilities, significantly outperform other algorithms in eliminating temporal redundancy and preserving the essential features of disease progression. Further analysis utilizing SHAP interpretability confirmed that the health index and photochemical reflection index are pivotal in the model's decision-making process. Crucially, the optimized classification framework achieved robust performance, with the random forest model combined with simulated annealing feature selection attaining the highest overall accuracy of 0.793 and a Kappa coefficient of 0.597. Furthermore, models using traditional and newly constructed vegetation indices yielded strong early warning capabilities, achieving overall accuracies up to 0.782 and 0.751 (with Kappa coefficients of 0.564 and 0.556, respectively) in distinguishing healthy, latent, and early symptomatic stages. This study illustrates that the comprehensive integration of temporal logic and physical driving features can markedly enhance the sensitivity and robustness of early warning systems for leaf blight, facilitating a transition from ""symptom-driven"" to ""mechanism-driven"" disease monitoring. This advancement offers substantial practical guidance for the intelligent and digital management of the walnut industry.","[""Du Nan"", ""Xia Qiuhao"", ""Li Zihao"", ""Guo Zhongzhong"", ""Gao Shan"", ""Zhang Qiang"", ""Wang Shiwei"", ""Zhang Zhenjun"", ""Jin Qiang"", ""Zhang Rui""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-08,10.1016/j.saa.2026.128895,"[""Journal Article""]","[""Juglans"", ""Plant Leaves"", ""Plant Diseases"", ""Neural Networks, Computer"", ""Random Forest"", ""Algorithms"", ""Support Vector Machine"", ""Hyperspectral Imaging"", ""Time Factors""]", 42868757,https://pubmed.ncbi.nlm.nih.gov/42868757/,Rapid and nondestructive prediction of chlorophyll fluorescence parameter Fv/Fm in Juncao leaves based on Vis-NIR spectroscopy and deep transfer learning.,"The maximum quantum yield of photosystem II photochemistry (Fv/Fm) is an important chlorophyll fluorescence parameter for evaluating plant photosynthetic activity and stress responses. Reliable Fv/Fm measurement using pulse amplitude modulated fluorometry generally requires approximately 30 min of dark adaptation, which reduces measurement throughput when large numbers of samples need to be evaluated. In this study, a rapid and nondestructive method was developed for predicting leaf Fv/Fm in Juncao seedlings by integrating visible-near infrared (Vis-NIR) spectroscopy, deep learning, and transfer learning. Vis-NIR spectra (400-1075 nm), SPAD, and Fv/Fm data were collected under different temperature treatments. Raw and preprocessed spectra were reconstructed into two-dimensional pseudo-images and used as inputs to an improved 18-layer residual network incorporating a convolutional block attention module (ResNet18-CBAM). The model was compared with conventional machine learning methods, and SHapley Additive exPlanations (SHAP) were used for interpretation. Transfer learning was performed using Jujuncao as the source domain and king grass as the target domain. The results showed that the ResNet18-CBAM model based on pseudo-images reconstructed directly from the raw spectra achieved the best prediction performance, with a prediction set coefficient of determination (RP2) of 0.9703, a root mean square error of prediction (RMSEP) of 0.0242, and a residual predictive deviation (RPD) of 5.8269. SHAP analysis revealed that the most influential wavelengths were mainly associated with photosynthetic pigment absorption, chlorophyll fluorescence emission, lipid related responses, and leaf water status. For the target domain with 120 samples, progressive fine-tuning using candidate samples selected by the Kennard-Stone algorithm showed that the cumulative subset containing 36 samples achieved the highest validation performance under the current sampling protocol, yielding a validation coefficient of determination (RV2) of 0.8154, representing an improvement of 19.06 percentage points relative to the model without target domain fine-tuning. These results demonstrate that integrating Vis-NIR spectroscopy with deep transfer learning enabled rapid and nondestructive Fv/Fm prediction across Juncao species, offering an effective approach for rapid assessment and high-throughput phenotyping of photosynthetic activity in Juncao and other bioenergy crops.","[""Xu Jinchai"", ""Zheng Conghui"", ""Lin Jiahui"", ""Pan Shuaibin"", ""Bian Shengtai"", ""Liu Shihan"", ""Xiao Jiajing"", ""Huang Zhenxiong"", ""Ye Dapeng"", ""Kong Xiangzeng"", ""Weng Haiyong""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-09,10.1016/j.saa.2026.128945,"[""Journal Article""]","[""Plant Leaves"", ""Chlorophyll"", ""Spectroscopy, Near-Infrared"", ""Fluorescence"", ""Deep Learning"", ""Photosystem II Protein Complex"", ""Photosynthesis""]", 42854524,https://pubmed.ncbi.nlm.nih.gov/42854524/,Machine learning-enabled fluorescence spectrum source tracing of typical pollutants in industrial wastewater.,"Complex and difficult-to-trace pollutants in industrial wastewater are posing major challenges for regulation and treatment. The known source identification methods rely on rough water quality data and models, which limits the accuracy of source tracing. In this study, a convolutional neural network-principal component analysis-random forest (CNN-PCA-RF) fusion framework that directly utilizes the three-dimensional fluorescence excitation-emission matrix (EEM) data was established. An EEM database with five sectors (chemical, smelting, fine chemical, pesticide manufacturing, biotechnology) and 207 datasets was constructed. The 2D-CNN extracts spatial-spectral EEM features, PCA performs dimensionality reduction, and RF provides stable classification. Under a multi-concentration data training setting, identification accuracies reached 95.99% at the outlet level and 96.85% at the sector level. Under validation with previously unseen random dilution factors, accuracies were 90.35% and 97.37%, respectively.","[""Xu Mingwang"", ""Yang Zhe"", ""Hao Yan"", ""Yu Huibin"", ""Dai Yunrong"", ""Yin Lifeng""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-07,10.1016/j.saa.2026.128942,"[""Journal Article""]",, 42854522,https://pubmed.ncbi.nlm.nih.gov/42854522/,Machine-learning-assisted retrieval of four temperatures from CO2 nonequilibrium transmittance spectra.,"Accurate retrieval of rotational and vibrational temperatures from non-equilibrium CO2 spectra is essential for diagnosing discharge plasmas, monitoring combustion flows, and modeling Mars entry aerothermodynamics. However, retrieving multiple temperature parameters from CO2 transmittance spectra constitutes a strongly nonlinear and ill-posed inverse problem. Conventional iterative spectral fitting methods are computationally expensive, sensitive to initial guesses, and prone to local-minimum trapping, making stable four-temperature retrieval challenging. To address these limitations, a non-iterative surrogate inverse framework based on a multilayer perceptron (MLP) is proposed for direct mapping from CO2 transmittance spectra to four-temperature states. A large-scale database containing 810,000 spectrum-temperature pairs was constructed using a line-by-line forward model based on RADIS and the Treanor non-equilibrium distribution. The input consists of 401-dimensional transmittance spectra in the 2284.2-2284.6 cm-1 range, while the outputs are the rotational temperature Trot and three vibrational temperatures Tv1, Tv2, and Tv3. Under noise-free conditions, the proposed model achieved mean absolute errors of 4.1 K, 7.5 K, 5.1 K, and 13.6 K, respectively, with all R2 values exceeding 0.998. Even under 3% Gaussian noise, the R2 values remained above 0.996. As a proof-of-concept demonstration, experimental validation on a CO2 discharge spectrum shows that the MLP retrievals agree with RADIS literature values (RMSE = 0.0387). Notably, the MLP method reduces the average retrieval time from 68 s to 0.012 s per spectrum, achieving a speedup of approximately 5700 times. This work is presented as an application study that addresses a specific diagnostic bottleneck - real-time four-temperature retrieval.","[""Ma Hao"", ""Bai Lu"", ""Li Jinlu""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-03,10.1016/j.saa.2026.128913,"[""Journal Article""]",, 42854521,https://pubmed.ncbi.nlm.nih.gov/42854521/,Simulation-based integration of an advanced spectroscopic refractive index biosensor via machine learning algorithms for detection of different concentrations of sucrose.,"Sucrose amounts in a water solution have diverse applications in healthcare, metabolic regulation, and food quality control. It is widely utilized as an alternative for physiological circumstances, fermenting phases, or quality assurance based on its level of concentration in biological and industrial samples. Current detection methodologies, such as enzymatic tests or chromatography, provide remarkable precision but tend to be time-consuming, dependent on reagents, and insufficient for monitoring in real time. To fix these limitations, we have envisioned a Four-Arm Ring-Embedded Refractive Index Sensor (FARERIS) to determine various concentrations of sucrose. Additionally, this study has been elaborated with a machine learning approach, which gives accurate as well as time-saving results. The design of the geometry of a quad-arm pattern leads to multiple pathways of light coupling while simultaneously enhancing interaction between the concentrating optical structures and their surroundings, the analyte. The achieved significant sensitivity values of 1553.39 nm/RIU, 1600.00 nm/RIU, 1518.98 nm/RIU, 1481.48 nm/RIU, 1647.05 nm/RIU, 1477.27 nm/RIU, and significant values for detection limits are 0.000269 RIU, 0.000440, 0.000310, 0.000434, 0.000360, and 0.000332 for the sucrose concentration of 5%, 10%, 15%, 20%, 25%, and 30%, correspondingly. The significant quality factor of 1723.96, and the significant figure of merit of 1283.79 RIU-1 have been obtained for detecting 5% of sucrose concentration, with the determined significant value of neural network fitting is 0.96475. The fusion of optical sensing and computational intelligence serves as a feasible approach towards advanced, adaptable biosensor devices.","[""Kamani Trupti"", ""Patel Shobhit K"", ""Sharma Yogesh"", ""Al-Zahrani Fahad Ahmed""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-06,10.1016/j.saa.2026.128943,"[""Journal Article""]","[""Sucrose"", ""Machine Learning"", ""Biosensing Techniques"", ""Refractometry"", ""Limit of Detection"", ""Algorithms"", ""Spectrum Analysis"", ""Computer Simulation""]", 42849215,https://pubmed.ncbi.nlm.nih.gov/42849215/,"DFT-assisted Raman spectroscopic characterization and machine learning classification of three rice types from the Taihu Lake Basin, China.","A DFT-assisted Raman framework integrating difference-derivative-PCA highly distinguishable Raman bands screening and difference visualization (DD-PCA-HSV), characteristic peak correlation analysis, and machine learning classification was developed for three milled rice types (SXJ-100, FYXZ, and BJR) from the Taihu Lake Basin. Five independent lots were analyzed for each type. Lots L1-L4 formed the development dataset, and L5 was reserved for independent testing. DFT calculations on simplified amylose and amylopectin models, together with literature, supported the assignment of 22 characteristic Raman peaks. Grouped four-fold cross-validation of DD-PCA-HSV on L1-L4 reproducibly retained six bands at 439, 576, 939, 1126, 1262, and 1460 cm-1. Z scores of these bands characterized spectral differences among the three rice types in the development dataset, and the same relative trends were reproduced in L5. Differences between BJR and SXJ-100/FYXZ were mainly associated with pyranose-ring and glycosidic-linkage vibrations, whereas differences between SXJ-100 and FYXZ were mainly associated with CO and CCO stretching vibrations. Linear correlation analysis combined with a dual-threshold rule identified 1080/1106 and 524/615 cm-1 as discriminative peak pairs. Subsampling analysis indicated that 30 spectra were sufficient for repeatable R2 estimation. The corresponding linear correlation features were reproduced in L5. Among RF, DT, and SVM models, SVM using normalized spectra achieved the best independent test performance, with accuracy, macro-precision, macro-recall, and macro-F1 all reaching 0.98. Overall, the framework distinguished the investigated rice types while linking discriminative Raman features to molecular vibrations, thereby providing a chemically interpretable basis for Raman-based rice characterization.","[""Hua Yisheng"", ""Chen Guoqing"", ""Li Mingzhe"", ""Meng Liangjie"", ""Zhao Rui"", ""Li Jie"", ""Ma Chaoqun"", ""Gao Hui"", ""Li Lei"", ""Yang Taiqun"", ""Zhu Chun"", ""Hu Anqi"", ""Shang Yunpeng"", ""Qian Yali""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-05,10.1016/j.saa.2026.128925,"[""Journal Article""]",, 42849212,https://pubmed.ncbi.nlm.nih.gov/42849212/,Quantification of palm oil adulteration in Sacha inchi oil: a comparative study between chemometrics and machine learning coupled with FTIR spectroscopy.,"The growing demand of Sacha inchi (Plukenetia volubilis) oil (SIO) has led to its high vulnerability to fraudulent economic adulteration with lower price alternatives, such as palm oil (PaO). As a consequence, analytical methods capable of authenticity verification are highly urgent. This study evaluated the effectiveness of Fourier transform infrared (FTIR) spectroscopy combined with chemometrics and machine learning algorithms to detect and quantify PaO adulteration in SIO. The chemometrics utilized partial least square (PLS) and principal component regression (PCR) optimized at various wavenumber ranges, whereas six advanced machine learning frameworks were performed. Among the models tested, the machine learning technique of k-nearest neighbors (kNN) algorithm emerged as the most powerful predictive model among others. The optimized kNN model achieved an exceptional goodness of fit, particularly at the wavenumber range of 3100-2800 cm-1 with R2 = 0.999, root mean square error (RMSE) of 0.756, and mean absolute error (MAE) of 0.528. Furthermore, the box-plot, diagnostic scale-location and residual plot analyses confirmed that kNN maintained strict homoscedasticity and eliminated systematic bias across the entire concentration levels of PaO. The ANOVA test demonstrated the highly significant difference in model capabilities (p < 0.001). Moreover, the summary statistics confirmed that the kNN model achieved the highest mean accuracy for all wavenumber ranges used, with R2 of 0.9985 and the lowest mean accuracy of both RMSE (1.1258) and MAE (0.883). These findings highlighted that integrating FTIR spectroscopy with machine learning models offers a precise, reliable, and powerful analytical tool for quality control and ensuring regulatory compliance of high-value edible oils. Further validation using diverse commercial and market samples is required to confirm the robustness, generalizability, and practical applicability of the developed models.","[""Windarsih Anjar"", ""Harumningtyas Anjar Anggraini"", ""Jatmiko Tri Hadi"", ""Wheni Indrianingsih Anastasia"", ""Anggraeni Ayu Septi"", ""Arisanti Anisa Galuh"", ""Utami Indrawati Dian"", ""Alam Lucky Prabowo Miftachul"", ""Rohman Abdul"", ""Lukitaningsih Endang""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-02,10.1016/j.saa.2026.128918,"[""Journal Article"", ""Comparative Study""]","[""Spectroscopy, Fourier Transform Infrared"", ""Palm Oil"", ""Machine Learning"", ""Least-Squares Analysis"", ""Principal Component Analysis"", ""Food Contamination"", ""Arecaceae"", ""Chemometrics"", ""Algorithms"", ""Plant Oils""]", 42849209,https://pubmed.ncbi.nlm.nih.gov/42849209/,Near-infrared spectroscopy-based parallel assessment of soybean fungal contamination and protein content using an adaptive-threshold Monte Carlo UVE strategy.,"Soybean protein content and fungal contamination level are critical determinants of economic value and consumption safety. However, conventional analytical methods rely on destructive chemical analyses or single-parameter measurements, severely limiting assessment efficiency and multi-attribute evaluation. Although near-infrared spectroscopy (NIRS) is widely used, it frequently suffers from spectral redundancy, model instability, and limited generalization in complex agricultural matrices. To address these challenges, this study proposes a parallel NIR analytical framework driven by Ensemble Monte-Carlo Stability-based Uninformative Variable Elimination (EMS-UVE). EMS-UVE integrates repeated stability estimation with a noise-referenced adaptive threshold for high-efficiency informative wavelength selection. For fungal contamination classification, EMS-UVE-XGBoost reduced the feature subset to 12 wavelengths while improving the validation accuracy and macro-F1 score to 0.9529 and 0.9530, respectively, outperforming the baseline ensemble Monte Carlo uninformative variable elimination-based extreme gradient boosting (EMCUVE-XGBoost) (21 wavelengths; accuracy = 0.9412, macro-F1 = 0.9418). For protein content regression, EMCUVE-SVR retained 18 wavelengths, with Rcv=0.8834, RMSECV = 1.6027%, Rp=0.8875, and RMSEP = 1.3338%, whereas EMS-UVE-SVR retained 13 wavelengths, achieving superior performance (Rcv=0.9012, RMSECV = 1.4720%, Rp=0.9070, and RMSEP = 1.2042%). Across 50 independent runs, EMS-UVE-XGBoost yielded a mean validation accuracy of 0.9518 ± 0.0062 and a macro-F1 score of 0.9516 ± 0.0068, while EMS-UVE-SVR achieved af (Rp) of 0.9055 ± 0.0108 with an RMSEP of 1.219 ± 0.058%, demonstrating exceptionally low run-to-run variability. Furthermore, the finalized dual-task models were successfully deployed on a Raspberry Pi 5-based embedded platform. External validation confirmed that the system completed shared spectral acquisition and simultaneous multi-attribute inference within an end-to-end measurement cycle of approximately 5.6 s, achieving a classification accuracy of 0.9200 and a protein prediction correlation coefficient of 0.9067. These findings demonstrate an effective, robust, and edge-deployable strategy for rapid, non-destructive, and multi-attribute evaluation of complex agricultural products.","[""Song Hao-Ran"", ""Qiao Jin-Hua"", ""Cao Jia-Lu"", ""Kang Ding-Rong"", ""You Chun-Xu"", ""Wu Yi-Teng"", ""Zhang Qi-Yuan"", ""Kamruzzaman Mohammed"", ""Barbin Douglas F"", ""Su Wen-Hao""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-10-02,10.1016/j.saa.2026.128912,"[""Journal Article""]","[""Spectroscopy, Near-Infrared"", ""Monte Carlo Method"", ""Glycine max"", ""Soybean Proteins"", ""Fungi"", ""Food Contamination"", ""Boosting Machine Learning Algorithms""]", 42800255,https://pubmed.ncbi.nlm.nih.gov/42800255/,Machine learning-enabled multivariate optimization and model-inferred process associations in the co-precipitation synthesis of Y2O3 nanoparticles.,"The co-precipitation synthesis of yttrium oxide (Y2O3) nanoparticles involves coupled precursor, acidity, dispersant, and thermal variables. We developed a machine-learning workflow to model average particle size from 30 MaxMin-sampled experiments in a six-dimensional process space. Nine regression algorithms were compared using leave-one-out prediction vectors with Bayesian hyperparameter optimization. The selected support vector regression model gave R2 = 0.83, RMSE = 20.82 nm, MAE = 16.63 nm, and MedAE = 15.38 nm. SHAP and LIME identified model-level associations among synthesis variables, while follow-up DLS, zeta-potential, conductivity, FTIR, and TG measurements were consistent with a tentative SDBS/PEG6000 precursor-state regulation hypothesis. Physically constrained virtual screening generated six experimentally tested G2 routes. Their measured particle sizes ranged from 35.4 to 63.2 nm, and one G2-S05 batch yielded phase-pure, quasi-spherical Y2O3 nanoparticles with a TEM-derived particle-level size distribution of 35.4 ± 8.3 nm and a BET surface area of 32.79 m2·g-1. These results establish a proof-of-concept workflow for hypothesis generation and process selection in data-limited oxide synthesis.","[""Meng Han"", ""Chen Juan"", ""Zhang Yifan"", ""Su Yuqiao"", ""Ban Tao"", ""Lin Jing"", ""Tao Jinzhang"", ""Zhou Yefei"", ""Gao Hongyi""]",Journal of colloid and interface science,2026-09-25,10.1016/j.jcis.2026.141660,"[""Journal Article""]",, 42665118,https://pubmed.ncbi.nlm.nih.gov/42665118/,An AI-driven clinical decision support system for coronary artery disease prediction in T2DM patients using clinical laboratory parameters and deep fusion learning.,"This article presents an intelligent clinical decision support system that integrates graph-based modelling and deep fusion learning for coronary artery disease (CAD) risk prediction using a clinical Laboratory and validated dataset. A Graph-Based Multivariate Kernel Density Estimation (KDE) imputation approach is proposed to handle the missing values, while latent class analysis (LCA) is applied for outlier detection. The framework further incorporates with the graph-based directional risk scoring and the comorbidity network analysis to identify significant clinical attributes and capture complex relationships among risk factors. For prediction, the Enhanced TabNet and Enhanced Residual MLP (ResNet) architectures are then combined through the probabilistic fusion with XGBoost as a meta-learner. The proposed system achieves 96.77% accuracy, demonstrating improved robustness and predictive performance.","[""Zahedi Mehboob"", ""Dabla Pradeep Kumar"", ""Das Subham"", ""Kandar Shyamalendu""]",Clinica chimica acta; international journal of clinical chemistry,2026-08-28,10.1016/j.cca.2026.121292,"[""Journal Article""]","[""Humans"", ""Coronary Artery Disease"", ""Decision Support Systems, Clinical"", ""Diabetes Mellitus, Type 2"", ""Artificial Intelligence"", ""Deep Learning"", ""Boosting Machine Learning Algorithms"", ""Prediction Algorithms"", ""Predictive Learning Models""]", 42648526,https://pubmed.ncbi.nlm.nih.gov/42648526/,"RNA biosensors in oncology: Mechanisms, Cancer-specific applications, and a Hallmark-aligned clinical roadmap.","Cancer diagnosis continues to rely on invasive tissue sampling and static molecular assessments that cannot reflect the real time RNA alterations driving tumour progression. RNA biosensors are genetically encoded or synthetic devices that translate specific RNA markers and tumour microenvironment signals into measurable outputs offer a compelling alternative, particularly for liquid biopsy applications where non-invasive, dynamic monitoring is essential. This review systematically examines key RNA biosensor classes developed for oncology, spanning fluorescence based platforms such as aptamers, FRET probes, and molecular beacons, enzymatic and electrochemical architectures including CRISPR Cas systems and field effect transistors, metabolite responsive designs encompassing riboswitches, RNA thermometers, and reactive oxygen species sensors and sequence specific toehold switches alongside exosomal detectors. For each class, we discuss operating principles, cancer relevant applications, reported detection thresholds reaching into the zeptomolar range, and current limitations. These biosensor capabilities are mapped onto Hanahan's hallmarks of cancer, and practical clinical roadmaps are outlined for three priority applications early population screening, longitudinal therapy response and resistance monitoring, and tumour microenvironment prognostication. Convergence with microfluidic integration, AI assisted interpretation, and multiplexed nanotechnology represents the critical next step in translating these platforms from laboratory tools into routine diagnostic practice.","[""Gamage A"", ""Herath H M L P B"", ""de Silva K M Nalin"", ""de Silva Rohini M""]",Clinica chimica acta; international journal of clinical chemistry,2026-08-26,10.1016/j.cca.2026.121306,"[""Journal Article"", ""Review""]","[""Humans"", ""Biosensing Techniques"", ""Neoplasms"", ""RNA""]", 42648524,https://pubmed.ncbi.nlm.nih.gov/42648524/,Development and validation of a generalizable M-protein screening model using routine laboratory indicators: a multicenter retrospective study.,"BACKGROUND: Early detection of plasma cell disorders (PCDs) remains challenging due to limited accessibility of gold standard diagnostic methods. This study aimed to develop a simple M-protein screening model using routine laboratory indicators for clinical laboratories. METHODS: A total of 5217 participants from three Chinese hospitals were enrolled. The derivation cohort (n = 3019) was randomly divided into training and internal validation cohorts. Two external validation cohorts (n = 1747 and n = 451) were included. M-protein positivity was rigorously defined by SPE combined with IFE. Demographic data and routine laboratory blood parameters were collected for model development. Eight machine learning algorithms-Logistic regression (LR), k-nearest neighbors classifier, decision tree classifier, random forest classifier, AdaBoost classifier, linear discriminant analysis, quadratic discriminant analysis, and multilayer perceptron classifier-were used to construct M-protein screening models. RESULTS: Eight M-protein screening models were established, incorporating indicators including sex, age, total protein, albumin, albumin/globulin ratio, and hemoglobin. Based on a comprehensive evaluation of models' performance and generalization ability, the LR model was identified as the optimal model, with an AUC of 0.843. A five-tier risk stratification was established based on predicted probabilities (P): ≤15.0%, 15.0-40.0%, 40.0-70.0%, 70.0-90.0%, and ≥ 90.0%. The model's performance in the internal validation cohort and two external validation cohorts also met expectations, with AUCs of 0.843, 0.801, and 0.800, respectively. CONCLUSION: We developed and validated a practical M-protein screening model based on routine laboratory indicators. The LR model demonstrated robust predictive ability and offers an accessible tool to facilitate early identification of individuals at high risk of M-protein, thereby supporting early diagnosis of clinically relevant PCDs, particularly in resource-limited primary healthcare settings.","[""Luo Hou-Long"", ""Zhou Lanxin"", ""Xiao Hua"", ""Zhang Qinxin"", ""Li Xin"", ""Ye Jiahui"", ""Liu Youlin"", ""Li Jin"", ""Liang Yonggang"", ""Xu Anping"", ""Ji Ling""]",Clinica chimica acta; international journal of clinical chemistry,2026-08-26,10.1016/j.cca.2026.121307,"[""Journal Article"", ""Multicenter Study"", ""Observational Study"", ""Validation Study""]","[""Adult"", ""Female"", ""Humans"", ""Male"", ""Middle Aged"", ""Classification Algorithms"", ""Machine Learning"", ""Predictive Learning Models"", ""Retrospective Studies"", ""Neoplasms, Plasma Cell""]", 42641935,https://pubmed.ncbi.nlm.nih.gov/42641935/,MicroRNA profiles as diagnostic tools in pediatric acute lymphoblastic leukemia.,"Acute lymphoblastic leukemia (ALL) represents the most common malignancy diagnosed in children, accounting for approximately 25-30% of all pediatric cancers. Despite remarkable improvements in survival rates over recent decades, with cure rates exceeding 90% in developed countries, the diagnosis and monitoring of ALL continue to rely on invasive bone marrow aspiration and morphological assessment. MicroRNAs (miRNAs), small non-coding RNA molecules that regulate gene expression post-transcriptionally, have emerged as promising biomarkers for cancer diagnosis and prognosis due to their stability in biological fluids, tissue-specific expression patterns, and regulatory roles in oncogenesis. This comprehensive review systematically examines the current evidence regarding miRNA expression profiles in pediatric ALL, focusing on their diagnostic potential, prognostic value, and applicability as non-invasive biomarkers. Aberrant miRNA expression has been consistently demonstrated in pediatric ALL, with specific signatures distinguishing B-cell precursor ALL from T-cell ALL, identifying cytogenetic subtypes, and predicting treatment response. Circulating miRNAs, particularly those packaged within exosomes, offer non-invasive alternatives for minimal residual disease monitoring and early relapse detection. Machine learning approaches integrating multi-miRNA panels have achieved high diagnostic accuracy, with several miRNAs, including miR-128-3p, miR-181a, miR-196b, miR-200c, and miR-326, demonstrating significant clinical utility. MicroRNA profiling represents a transformative approach to pediatric ALL management, offering enhanced diagnostic precision, improved risk stratification, and non-invasive monitoring capabilities. While challenges remain in standardizing detection methodologies and establishing clinically validated thresholds, the integration of miRNA biomarkers into clinical practice holds substantial promise for personalized medicine in childhood leukemia.","[""Abuhassan Qamar"", ""Al-Khreisat Mutaz Jamal"", ""Younis Shahad Mohammed Dhiaa"", ""Kubaev Aziz"", ""Shaymardanov Olim"", ""Salih Rasim M"", ""Jaber Mohamed Adil""]",Clinica chimica acta; international journal of clinical chemistry,2026-08-25,10.1016/j.cca.2026.121299,"[""Journal Article"", ""Review""]","[""Humans"", ""MicroRNAs"", ""Precursor Cell Lymphoblastic Leukemia-Lymphoma"", ""Child"", ""Biomarkers, Tumor""]", 42624190,https://pubmed.ncbi.nlm.nih.gov/42624190/,Decision-making logic for samples of clinically significant EDTA-dependent pseudothrombocytopenia with machine learning.,"OBJECTIVES: To establish practical, clinically meaningful criteria for identifying significant platelet aggregation in suspected samples, defined by a platelet-count deviation exceeding the reference change value (RCV). METHODS: Suspected aggregated samples were collected. Venous blood anticoagulated with citrate/EDTA underwent platelet counting via impedance method (PLT-I) and blood smear analysis at 5-, 30-, 60-, and 120-min intervals. Manual microscopy (30-min smears) assessed aggregation cluster size/number, while digital cytomorphology (PLT-pro) quantified clusters at all time points. Using 5-min citrate platelet counts as reference, relative deviations (d%) of EDTA PLT-I were calculated. Clinically significant aggregation-positive samples were defined as d% exceeding the negative RCV (30.02% for PLT < 100 × 10(Zhang et al., 20259)/L; 20.92% for PLT ≥ 100 × 10(Zhang et al., 20259)/L). Samples were randomly split into training (70%) and test (30%) sets to develop logistic regression and decision-tree models. RESULTS: A total of 86 patients contributed 338 time-point results; 183 were classified as positive and 155 as negative. Manual microscopy showed that counting clusters with ≥5 platelets (vs. ≥ 3) improved specificity (p = 0.083). A criterion of ≥3 clusters, each containing ≥5 platelets, achieved 100% sensitivity and 33.3% specificity. PLT-I and PLT-pro were key discriminators via the Boruta algorithm. In the test set, the decision-tree model outperformed logistic regression (sensitivity: 92.42% vs. 89.39%; specificity: 94.29% vs. 94.29%; accuracy: 93.07% vs. 91.09%). CONCLUSION: Criteria were established for manual microscopy (≥ 3 clusters with ≥5 platelets) and digital analysis (PLT-pro 27-97 with PLT-I < 58 × 10(Zhang et al., 20259)/L, or PLT-pro ≥97 with PLT-I < 164 × 10(Zhang et al., 20259)/L). However, these thresholds are merely exploratory results, and independent validation must be performed in this laboratory before they can be used for clinical decision-making.","[""Meng Xin"", ""Li Na"", ""Yang Liu"", ""Wang Xiaoqin"", ""Wu Wenjing""]",Clinica chimica acta; international journal of clinical chemistry,2026-08-20,10.1016/j.cca.2026.121291,"[""Journal Article""]","[""Humans"", ""Edetic Acid"", ""Pseudothrombocytopenia"", ""Machine Learning"", ""Platelet Aggregation"", ""Platelet Count"", ""Decision Making"", ""Blood Platelets""]", 42486315,https://pubmed.ncbi.nlm.nih.gov/42486315/,Automated detection of intravenous fluid contamination using longitudinal routine laboratory data: development and external validation of a multitask learning framework.,"OBJECTIVES: To develop and externally validate an automated framework for intravenous fluid contamination (IFC) screening using routinely available longitudinal laboratory data. METHODS: A bicentric retrospective study included an internal cohort (IC; Hospital La Fe, València) for model development and an external validation cohort (EVC; Hospital General Universitari de València) composed of hospitalized patients with a previous determination within 5 days. Normalized relative change variables were calculated from paired measurements. Controls were retrospectively selected in the IC after removal of outliers using an autoencoder; contaminated samples were generated through controlled in vitro admixture with normal saline and glucose-containing solutions. A multitask neural network provided contamination probability, likely contaminant type, and severity-oriented estimation. Performance was evaluated using repeated internal 80:20 partitions (IRP; n = 30), the EVC, and an exploratory real-world dataset of retrospectively suspected IFC cases (RW-IFC). Model outputs were used to derive a stratified risk framework. RESULTS: The IC included 1760 samples (50% IFC), the EVC 1371 samples (5.4% IFC), and RW-IFC 37 samples. Discriminative performance remained high (ROC-AUC 0.94-0.96), with sensitivities and specificities ≥85%. Most false-negatives corresponded to contamination ≤5%, and the mean predicted contamination level among false-positive cases was 2.12 ± 1.06%. Multiclass accuracy ranged 0.87-0.90. Severity estimation showed strong internal correlation (ρ = 0.83; p < 0.0001) and good external agreement (0.84 within ±1 level), although correlation decreased in the EVC. 97% of RW-IFC cases were correctly identified. Risk stratification in the EVC achieved a sensitivity of 99.1% while reducing unnecessary alerts by 17%. CONCLUSIONS: This framework showed potential to support automated IFC screening.","[""Tortosa-Carreres Jordi"", ""Martí-Montoro Nahúm"", ""Acevedo-Galvis Jonnathan Andrés"", ""Pascual-Escrivà Cristel"", ""Alonso-Díaz Ricardo"", ""Fuster-Lluch Óscar"", ""Sahuquillo-Frias Laura"", ""Laiz-Marro Begoña""]",Clinica chimica acta; international journal of clinical chemistry,2026-07-22,10.1016/j.cca.2026.121242,"[""Journal Article"", ""Validation Study""]","[""Humans"", ""Retrospective Studies"", ""Automation"", ""Neural Networks, Computer"", ""Machine Learning"", ""Longitudinal Studies""]", 42476280,https://pubmed.ncbi.nlm.nih.gov/42476280/,"Precision forensic toxicology: Leveraging Toxicogenomics, multi-omics, and artificial intelligence for next-generation poisoning investigations.","For much of its history, forensic toxicology was built on a core analytical question: what poison or drug is present, and at what concentration? That question still matters, but modern casework increasingly shows that it is no longer sufficient on its own. The field grew from the nineteenth-century chemical toxicology of Mathieu Orfila and the arsenic-detecting Marsh test into a discipline central to medicolegal death investigation, overdose certification, and exposure reconstruction. Yet contemporary practice must now contend with novel psychoactive substances, highly potent synthetic opioids, complex polysubstance deaths, delayed reporting, postmortem redistribution, matrix degradation, and incomplete toxicological context. These pressures are pushing forensic toxicology beyond simple compound detection toward mechanism-aware, data-rich interpretation. This review develops the idea of precision forensic toxicology as a proposed, forward-looking conceptual framework for that transition rather than a description of established or routinely implemented forensic practice. In this model, classical analytical chemistry remains essential, but it is connected to toxicogenomics, transcriptomics, epigenomics, proteomics, metabolomics, exposomics, microbiomics, and spatial multi-omics, then integrated through systems toxicology and artificial intelligence. Recent literature shows that toxicogenomics can reveal early molecular perturbations before overt phenotype, that postmortem metabolomics can assist cause-of-death screening, that RNA- and miRNA-based markers may improve postmortem interval estimation, and that machine learning can strengthen high-resolution mass-spectrometry workflows and multi-omics interpretation. At the same time, routine implementation remains constrained by standardization, validation, explainability, privacy governance, and regulatory uncertainty. None of these components is yet part of routine forensic casework; each remains at a research or early-validation stage, and their translation into daily practice will require dedicated feasibility, validation, and regulatory work, which this review discusses explicitly. Taken together, the evidence suggests that the future of poisoning investigation will depend less on any single assay and more on interoperable workflows that connect toxicant detection with biological response, individual susceptibility, and transparent decision support, and that these workflows are intended to complement, not replace, classical toxicological and medicolegal interpretation.","[""Nurkolis Fahrul"", ""Amri Muhammad Fadli"", ""Sibarani Jeremy Nicolas"", ""Tjandrawinata Raymond Rubianto""]",Clinica chimica acta; international journal of clinical chemistry,2026-07-20,10.1016/j.cca.2026.121239,"[""Journal Article"", ""Review""]","[""Humans"", ""Forensic Toxicology"", ""Toxicogenetics"", ""Artificial Intelligence"", ""Multiomics"", ""Poisoning""]", 42331130,https://pubmed.ncbi.nlm.nih.gov/42331130/,Establishment and validation of intelligent review rules for blood cell analysis based on the gradient boosting decision tree algorithm.,"OBJECTIVE: To develop artificial intelligence (AI)-based intelligent review rules to accurately screen samples that require retests or blood smear microscopy examinations, thereby reducing the review rate, improving the work efficiency of haematology laboratories, and ensuring the quality of whole blood analysis. METHODS: A total of 10,212 EDTA-K2 venous blood samples from the clinical laboratories of 4 hospitals from May 2022 to April 2025 were collected. Among them, 9000 samples from the First People's Hospital of Foshan, the Zhongda Hospital affiliated to Southeast University, and the First Affiliated Hospital of Sun Yat-sen University were used to establish the rules, and 1212 samples from the Zhongshan Hospital affiliated to Fudan University were used for rule validation. The obtained rules were compared with the 41 review rules set by the International Consensus Group for Haematology Review (ICGHR). All samples were tested using a haematology analyser under the customized dilution ratio (CDR) mode and smear microscopy, if the microscopic examination results trigger the 10 microscopic anomalous rules established by the ICGHR, then the sample is deemed to be an anomalous sample. Intelligent review rules were established using the gradient boosting decision tree (GBDT) algorithm. Moreover, the false negative rate, false positive rate, and review rate of the validation set were statistically analysed. RESULTS: (1) Among the 9000 samples in the establishment set, 4493 anomalous samples were identified by microscopic examination, and 26 intelligent review rules were established using the GBDT algorithm. (2) For the external validation set, 180 anomalous samples were identified by microscopic examination. Compared with those of the 41 international review rules, the false positive rate (14.60%) and review rate (27.55%) of the intelligent review rules were significantly lower (false positive rate: 30.19%, review rate: 44.14%). The false negative rate of the intelligent review rules (1.89%) was comparable to that of the 41 international review rules (0.90%) and met the ICGHR requirement of a false negative rate < 5%, with no critical hematologic cells missed. CONCLUSION: Intelligent review rules generated using AI techniques effectively reduce the false positive rate and review rate and prevent missed diagnoses of hematologic cells while ensuring test quality. The review rules generated by AI methods significantly increase laboratory efficiency and are especially suitable for medical institutions.","[""Tang Wenjia"", ""Chen Shaoqian"", ""Yu Zhenglin"", ""Gu Xiumei"", ""Pan Baishen"", ""Wang Beili"", ""Ding Rong"", ""Zhang Shihong"", ""Wang Chong"", ""Lin Jing"", ""Guo Wei""]",Clinica chimica acta; international journal of clinical chemistry,2026-06-22,10.1016/j.cca.2026.121196,"[""Journal Article"", ""Validation Study""]","[""Humans"", ""Decision Trees"", ""Boosting Machine Learning Algorithms"", ""Blood Cells"", ""Artificial Intelligence""]", 42957107,https://pubmed.ncbi.nlm.nih.gov/42957107/,"Integrative network pharmacology, machine learning, and experimental validation identify the TLR4/NF-κB/NLRP3 axis as a potential mechanism underlying the therapeutic effects of Lingguizhugan decoction in Alzheimer's disease.","ETHNOPHARMACOLOGICAL RELEVANCE: Lingguizhugan decoction, a traditional formula originating from Zhang Zhongjing's Treatise on Cold Damage and Miscellaneous Diseases, has shown cognitive benefits in APP/PS1 mice in our earlier study, partly through restoration of meningeal lymphatic drainage and enhancement of amyloid-β (Aβ) removal. Whether other molecular events contribute to its therapeutic action in Alzheimer's disease (AD) remains to be clarified. AIM OF THE STUDY: This work was designed to clarify additional potential mechanisms by which LGZG may act against AD. METHODS AND MATERIALS: The chemical constituents of LGZG were profiled by UPLC-Q-TOF-MS. Candidate targets and pathways were screened by integrating public databases, transcriptomic analysis, network pharmacology, and enrichment analysis. Eight machine learning models and SHAP analysis were subsequently used to prioritize candidate regulators. Molecular docking and molecular dynamics simulations were performed to evaluate the potential interactions between representative LGZG constituents and TLR4. APP/PS1 mice and BV2 microglial cells were used for experimental validation. Spatial learning and memory, hippocampal pathological alterations, Aβ deposition, inflammatory cytokines, microglial inflammatory phenotype-associated markers, and TLR4/NF-κB/NLRP3 signaling were evaluated using behavioral testing, histological staining, ELISA, immunofluorescence, qRT-PCR, and Western blotting. TAK-242 inhibition and TLR4 overexpression were further used to assess the involvement of TLR4 signaling. RESULTS: Twenty-three candidate constituents were retained following chemical profiling and in silico screening of LGZG. Integrated analysis of network pharmacology, AD-related genes, and transcriptomic data identified 46 candidate LGZG-AD-DEG targets. Machine learning and SHAP analyses further prioritized several inflammation-related regulators, including RELA and NFKBIA, supporting inflammation-related signaling as a major direction for subsequent validation. Molecular docking and molecular dynamics simulations suggested potential interactions between representative LGZG constituents and the TLR4 receptor. In APP/PS1 mice, LGZG improved spatial learning and memory, attenuated hippocampal neuronal injury, reduced Aβ deposition, and decreased IL-1β, IL-6, and TNF-α levels. LGZG also reduced hippocampal TLR4, P-IκBα, P-NF-κB, and NLRP3 expression and decreased TLR4/NF-κB/NLRP3-related signals in Iba1-positive microglia. In LPS-stimulated BV2 cells, LGZG shifted inflammatory phenotype-associated markers toward a less pro-inflammatory profile and suppressed TLR4/NF-κB/NLRP3 signaling. TAK-242 treatment and TLR4 overexpression experiments further supported the involvement of TLR4 signaling in these effects. CONCLUSION: LGZG alleviated cognitive impairment, hippocampal pathological alterations, Aβ deposition, and neuroinflammatory responses in APP/PS1 mice. Integrated bioinformatic analyses and experimental validation suggest that regulation of TLR4/NF-κB/NLRP3 signaling contributes to the anti-neuroinflammatory effects of LGZG in AD.","[""Hu Dezhi"", ""Li Deyu"", ""He Lin"", ""Mao Shiwei"", ""Zhu Yu-An"", ""Chen Xiaojing"", ""Zhang Jingyi"", ""Zhou Chunxiang"", ""Zhou Xibin""]",Journal of ethnopharmacology,2026-10-29,10.1016/j.jep.2026.122517,"[""Journal Article""]","[""Animals"", ""Toll-Like Receptor 4"", ""Alzheimer Disease"", ""NLR Family, Pyrin Domain-Containing 3 Protein"", ""Machine Learning"", ""Network Pharmacology"", ""NF-kappa B"", ""Drugs, Chinese Herbal"", ""Mice"", ""Molecular Docking Simulation"", ""Male"", ""Mice, Transgenic"", ""Signal Transduction"", ""Hippocampus"", ""Mice, Inbred C57BL"", ""Disease Models, Animal"", ""Molecular Dynamics Simulation"", ""Microglia""]", 42946182,https://pubmed.ncbi.nlm.nih.gov/42946182/,Integrated analysis of the blood-absorbed and gastric tissue-distributed components of Huangqi Fuling Decoction reveals the inhibitory effect on gastric cancer liver metastasis via the PRMT1/FOXO1 axis.,"ETHNOPHARMACOLOGICAL RELEVANCE: Huangqi Fuling Decoction (HQFLT) is a classic formula that has shown therapeutic potential against gastric cancer (GC) in clinical practice. However, its components and underlying mechanisms remain unclear. AIM OF THE STUDY: This study aimed to investigate the inhibitory effect of HQFLT on liver metastasis of GC and to elucidate its potential pharmacological basis and underlying mechanisms. MATERIALS AND METHODS: UPLC-MS was used to detect the absorbed components in blood and gastric tissues from mice administered intragastrically with HQFLT. WGCNA, LASSO, SVM-RFE, and random forest were applied to screen for HQFLT targets against GC, and the core target was validated in patient tissues. Downstream proteins interacting with the core target were predicted by the STRING database, and the interactions were further analyzed by molecular dynamics simulations. CCK-8, wound healing, Transwell, and rescue assays were performed to evaluate the effects of HQFLT on GC cells. In vivo validation was conducted using a mouse hematogenous colonization model with bioluminescence imaging, H&E, and IHC staining. RESULTS: UPLC-MS identified 128 components in blood and 188 in gastric tissues, with 65 common components of HQFLT detected in both blood and gastric tissues. Machine learning identified protein arginine methyltransferase 1 (PRMT1) as a core target of HQFLT, with elevated expression in GC tissues, and Forkhead box protein O1 (FOXO1) as its downstream target. Inhibition of PRMT1 or FOXO1 suppressed cell migration and invasion, while rescue assays showed that HQFLT exerted its effects partially through regulating the PRMT1/FOXO1 axis. In vivo, HQFLT inhibited liver metastasis of GC, as evidenced by reduced bioluminescence signals, and counteracted the pro-metastatic effect of PRMT1 expression inducer. CONCLUSIONS: This study elucidated the potential pharmacological basis of HQFLT and suggested that its inhibitory effect on liver metastasis of GC was partially mediated through the PRMT1/FOXO1 axis.","[""Lu Doudou"", ""Jia Shumin"", ""Song Chuang"", ""Liu Wenjing"", ""Wang Ruinan"", ""Zhao Jianan"", ""Feng Juju"", ""Liu Yu"", ""Qu Haiyan"", ""Nan Yi"", ""Yuan Ling""]",Journal of ethnopharmacology,2026-10-28,10.1016/j.jep.2026.122560,"[""Journal Article""]","[""Animals"", ""Forkhead Box Protein O1"", ""Drugs, Chinese Herbal"", ""Protein-Arginine N-Methyltransferases"", ""Liver Neoplasms"", ""Humans"", ""Stomach Neoplasms"", ""Repressor Proteins"", ""Mice"", ""Cell Line, Tumor"", ""Male"", ""Antineoplastic Agents, Phytogenic"", ""Cell Movement"", ""Signal Transduction"", ""Mice, Inbred BALB C""]", 42838301,https://pubmed.ncbi.nlm.nih.gov/42838301/,"Combined network pharmacology, transcriptomics and machine learning suggests Danggui Buxue Tang may suppress oxidative stress and inflammation through Nrf2/HO-1 and IL-17 pathways in pulmonary fibrosis.","ETHNOPHARMACOLOGICAL RELEVANCE: Danggui Buxue Tang (DBT), a classical Chinese herbal formula of Astragalus membranaceus (Fisch.) Bunge and Angelica sinensis (Oliv.) Diels in a 5:1 ratio, has long been used for qi and blood deficiency. Its mechanism against pulmonary fibrosis (PF) remains unclear. AIM OF THE STUDY: This study examined how DBT treats silica-induced PF, combining network pharmacology, transcriptomics, machine learning and experimental validation. MATERIALS AND METHODS: We built a rat PF model by intratracheal silica instillation. Network pharmacology and weighted gene co-expression network analysis (WGCNA) identified candidate therapeutic targets. Four machine learning algorithms screened diagnostic signature genes (DSGs). We profiled immune infiltration with CIBERSORT, modeled compound-target binding by molecular docking, and confirmed key genes and pathways by Western blotting. RESULTS: DBT treatment significantly attenuated silica-induced lung pathology, evidenced by improved pulmonary function, reduced alveolitis and fibrosis scores, diminished collagen deposition, and downregulated expression of α-smooth muscle actin, collagen I, and fibronectin. Furthermore, DBT markedly suppressed inflammatory responses (decreasing TNF-α, IL-6, IL-1β, and TGF-β1 while elevating IL-10) and mitigated oxidative stress (lowering MDA levels while enhancing SOD, GSH-Px, and T-AOC activities). Bioinformatic analyses identified three core DSGs-MMP7, MMP13, and TNF-as candidate therapeutic targets. Enrichment analyses highlighted the nuclear factor erythroid 2-related factor 2 (Nrf2)/heme oxygenase-1 (HO-1) pathway and IL-17 signaling as the primary mechanistic routes. Immune profiling revealed that DBT's therapeutic effects correlate with specific shifts in immune cell subsets. Ultra-performance liquid chromatography-quadrupole-orbitrap mass spectrometry detected 11 prototype constituents in serum; notably, the Ononin, Methylnissolin-3-O-glucoside, and Isomucronulatol-7-O-glucoside exhibited strong binding affinities to MMP7, MMP13, and TNF. DBT inhibited the profibrotic IL-17 cascade while activating the antioxidant Nrf2/HO-1 axis. CONCLUSION: Our data suggest that DBT may ease silica-induced PF through multiple targets that act on oxidative stress and inflammation. The DSGs (MMP7, MMP13, TNF) and pathways (Nrf2/HO-1, IL-17) help explain how DBT works and point to new targets for treating PF. The approach also provides a reusable workflow for studying other complex herbal formulas in fibrosis.","[""Yang Zhifeng"", ""Chen Yanqin"", ""Wen Chengli"", ""Chang Yuqian"", ""Zhang Haidong""]",Journal of ethnopharmacology,2026-10-06,10.1016/j.jep.2026.122476,"[""Journal Article""]","[""Animals"", ""Oxidative Stress"", ""Drugs, Chinese Herbal"", ""NF-E2-Related Factor 2"", ""Network Pharmacology"", ""Male"", ""Pulmonary Fibrosis"", ""Interleukin-17"", ""Rats, Sprague-Dawley"", ""Machine Learning"", ""Rats"", ""Molecular Docking Simulation"", ""Signal Transduction"", ""Transcriptome"", ""Inflammation"", ""Heme Oxygenase (Decyclizing)"", ""Heme Oxygenase-1"", ""Anti-Inflammatory Agents"", ""Disease Models, Animal"", ""Lung""]", 42810458,https://pubmed.ncbi.nlm.nih.gov/42810458/,"Total saponins from Paris polyphylla against breast cancer: an integrative pharmacological study combining chemical profiling, network pharmacology, machine learning, and in vitro validation.","ETHNOPHARMACOLOGICAL RELEVANCE: Paris polyphylla Smith var. yunnanensis (Franch.) Hand.-Mazz., a traditional heat-clearing and detoxifying Chinese herb, has been extensively used in cancer therapy as a key constituent of Chinese patent medicines. Although the antitumor activities of individual steroidal saponin monomers have been extensively investigated, the therapeutic potential and molecular targets of the total saponins from Paris polyphylla var. yunnanensis (TPP) in breast cancer have not yet been reported. AIM OF THE STUDY: This study aimed to elucidate the anti-breast cancer efficacy and underlying mechanisms of TPP by integrating chemical profiling, network pharmacology, bioinformatics, machine learning, molecular docking, and in vitro experimental validation. MATERIALS AND METHODS: The chemical composition of TPP was analyzed by UPLC-Q-TOF-MS and HPLC. Potential targets of TPP against breast cancer were predicted through network pharmacology and bioinformatics analyses, and further refined using protein-protein interaction network analysis combined with machine learning. The candidate targets were subsequently evaluated by ROC curve analysis, immune cell infiltration assessment, and molecular docking. Finally, the findings were validated through in vitro pharmacological assays, RT-qPCR, Western blotting and RNA transcriptomic sequencing (RNA-seq). RESULTS: Seventeen compounds were identified in TPP, with polyphyllin I, II, and VII constituting over 85% of the total active components. Through intersection analysis of TPP targets, WGCNA module genes, and DEGs, 111 potential therapeutic targets were identified. Subsequent protein-protein interaction network analysis and machinelearning prioritized four hub genes-LIPE, PPARG, FABP4, and IGF1-which were significantly downregulated in breast cancer tissues and validated with high diagnostic accuracy (AUC > 0.90). These genes were closely associated with tumor immune cell infiltration. Molecular docking predicted strong binding between the major saponins and key targets, and RT-qPCR and Western blotting further validated the reversal of their expression at both mRNA and protein levels. In vitro assays demonstrated that TPP significantly suppressed 4T1 cells proliferation, colony formation, and migration, while concurrently inducing apoptosis and pyroptosis. RNA-seq analysis further substantiated that TPP exerts its anti-breast cancer effects through modulation of lipid metabolic pathways. CONCLUSIONS: This study demonstrates that TPP exert anti-breast cancer effects by targeting the LIPE/PPARG/FABP4/IGF1 axis and triggering apoptosis and pyroptosis. These findings offer a foundation for ethnopharmacological application of TPP in breast cancer therapy and demonstrate the potential of integrative pharmacological strategies in developing targeted breast cancer treatment.","[""Huang Hua-Ting"", ""Zhou Li-Peng"", ""Chen Qi"", ""Wang Wen-Qi"", ""Yao Shu-Chang"", ""Zhang Hua-Juan"", ""Li Ya-Xuan"", ""Qu Chang-Hai"", ""Yin Xing-Bin"", ""Ni Jian"", ""Dong Xiao-Xv""]",Journal of ethnopharmacology,2026-09-29,10.1016/j.jep.2026.122453,"[""Journal Article""]","[""Saponins"", ""Network Pharmacology"", ""Machine Learning"", ""Female"", ""Humans"", ""Molecular Docking Simulation"", ""Antineoplastic Agents, Phytogenic"", ""Breast Neoplasms"", ""Melanthiaceae"", ""Cell Line, Tumor"", ""Animals"", ""Cell Proliferation"", ""Liliaceae""]", 42764122,https://pubmed.ncbi.nlm.nih.gov/42764122/,Baihe Ganmai Dazao Decoction ameliorates depression-like and sleep-disturbance phenotypes: potential involvement of IL-17A/IL-17RA-related inflammatory signaling and microglial polarization.,"ETHNOPHARMACOLOGICAL RELEVANCE: Originally documented in Zhang Zhongjing's Synopsis of Prescriptions of the Golden Chamber of the Eastern Han Dynasty, Ganmai Dazao Decoction (GMDZ) is composed of three medicinal materials: Glycyrrhizae Radix et Rhizoma (GRR), Tritici Levis Fructus (TLF), and Jujubae Fructus (JF). Traditionally, it has been used to treat zangzao, a syndrome characterized by emotional disturbance and sleep disorder, and is now widely prescribed for depression, anxiety, and insomnia. Nevertheless, the effective material foundation and intrinsic therapeutic mechanisms of GMDZ are still not thoroughly clarified. Based on GMDZ, Baihe Ganmai Dazao Decoction (BGMD) was developed by adding Lilii Bulbus (LB), Poria (PO), and Ziziphi Spinosae Semen (ZSS) to strengthen its Yin-nourishing and mind-calming effects. AIM OF THE STUDY: This study aimed to optimize the compatibility ratio of BGMD, elucidate its pharmacodynamic material basis and antidepressant and sleep-promoting mechanisms, and establish a quality control strategy. MATERIALS AND METHODS: The optimal compatibility ratio of BGMD was screened using a uniform design combined with the entropy weight method in zebrafish models of depression and sleep disturbance. To further characterize its therapeutic activities, a composite animal model coupling chronic unpredictable mild stress (CUMS) stimulation and 4-chloro-DL-phenylalanine (PCPA) administration was constructed to mimic depressive states alongside sleep dysfunction in mice. UPLC-Q-TOF-MS/MS, network pharmacology, bioinformatics, machine learning, and molecular docking were integrated to screen putative active constituents and key targets, followed by validation using RT-qPCR, immunohistochemistry, ELISA, and immunofluorescence. HPLC-based quantification of quality markers was performed, and the in vivo activities of candidate constituents were further verified in zebrafish. RESULTS: The optimal compatibility ratio of BGMD was TLF: GRR: JF: LB: PO: ZSS = 5: 3: 1: 6: 4: 2. BGMD effectively ameliorated depression-like behaviors and sleep dysfunction in model mice. It markedly upregulated the hippocampal contents of gamma-aminobutyric acid (GABA), dopamine (DA), 5-hydroxytryptamine (5-HT) and norepinephrine (NE), while reducing the levels of interleukin-1β (IL-1β) and corticosterone (CORT). BGMD corrected the imbalance in microglial M1/M2 phenotypes and alleviated neuronal damage within the CA1, CA3, and DG subregions of the hippocampus. Mechanistically, BGMD may exert its effects partly through modulation of IL-17A/IL-17RA-related inflammatory signaling and associated downstream signaling molecules, accompanied by suppression of Il17a, Il17ra, Traf6, p-STAT3, and p-JNK3. Eighty-six compounds were identified in BGMD, and eight of them were selected as candidate active constituents. Regaloside A, spinosin, and liquiritin were selected as representative quality-control indicator compounds, with mean contents of 0.0500, 0.0364, and 0.0844 mg/g, respectively. CONCLUSION: BGMD exerts synergistic antidepressant and sleep-promoting effects, which may be associated with modulation of IL-17A/IL-17RA-related inflammatory signaling, restoration of microglial polarization balance, protection of hippocampal neuronal integrity, and improvement of neurotransmitter homeostasis. These findings provide insights into the pharmacodynamic basis and potential mechanisms of BGMD and support its further quality-control standardization and pharmacological investigation.","[""Zhang Liyue"", ""Ren Haiqin"", ""Li Jianli"", ""Wang Jiafeng"", ""Zhao Junyi"", ""Zou Xiang"", ""Qu Zhongyuan""]",Journal of ethnopharmacology,2026-09-20,10.1016/j.jep.2026.122393,"[""Journal Article""]","[""Animals"", ""Drugs, Chinese Herbal"", ""Microglia"", ""Depression"", ""Male"", ""Sleep Wake Disorders"", ""Mice"", ""Signal Transduction"", ""Zebrafish"", ""Antidepressive Agents"", ""Phenotype"", ""Molecular Docking Simulation"", ""Disease Models, Animal""]", 42716194,https://pubmed.ncbi.nlm.nih.gov/42716194/,Exploring the molecular mechanisms of Weifuchun in the treatment of chronic atrophic gastritis based on single-cell sequencing and machine learning.,"ETHNOPHARMACOLOGICAL RELEVANCE: Chronic atrophic gastritis (CAG) is a common precancerous lesion of gastric cancer, for which there is currently a lack of effective drugs to reverse its pathological progression. Weifuchun (WFC), as a classical traditional Chinese medicine (TCM) compound, has demonstrated definitive clinical efficacy in treating CAG, although its molecular mechanisms remain unclear. AIM OF THE STUDY: To explore the potential molecular mechanisms of WFC in the treatment of CAG. MATERIALS AND METHODS: This study conducted single-cell RNA sequencing (scRNA-seq) on gastric mucosal tissues from patients with chronic superficial gastritis (CSG) and CAG, mapping gastric mucosal cell populations and screening differentially expressed genes. The CAG transcriptome data from six integrated databases from Gene Expression Omnibus (GEO) were combined with the databases from Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) and relevant literature to identify active components and potential targets of WFC, constructing a ""active components-cell populations-intersected target genes"" regulatory network. Eight machine learning models (including random forest) were employed to predict the efficacy of key target genes. Finally, a cell model was established using MNNG-induced GES-1 cells, with potential target protein expression validated by Western blot. RESULTS: Single-cell sequencing identified 17 cell types, with epithelial cells constituting the highest proportion (86.96%). In the CAG group, T/NK cells, plasma cells, and fibroblasts were increased as abserved, while goblet cells and enterocytes appeared exclusively in CAG samples. Pseudotime analysis revealed a preferential polarization of mucosal-associated invariant T (MAIT) cells toward a Th17- like phenotype in CAG, accompanied by a decline in IFNG+ effector CD8+ T cells. Cellular interaction analysis highlighted fibroblasts, macrophages, and dendritic cells as central hubs, with key ligand-receptor pairs (e.g., CCL5/CCR5, ADM/RAMP3) implicated in pro-inflammatory crosstalk. Thirteen active components of WFC were identified, and their target genes overlapped with CAG differentially expressed genes from single-cell sequencing, yielding five common targets (PDE4D, JUN, KCNH2, NFKBIA, MMP12). Further intersection with GEO- derived CAG DEGs identified JUN and MMP12 as potential targets. Machine learning demonstrated that the random forest model exhibited optimal classification performance (AUC = 1.000), with JUN being the candidate feature (single-gene discriminant accuracy approaching 95%). JUN, NFKBIA, and MMP12 formed the potential feature set. In vitro experiments showed that WFC concentration-dependently inhibited the proliferation of MNNG-induced GES-1 cells and significantly downregulated MMP12 protein expression (P < 0.05), while no significant effects were observed on JUN, NFKBIA, PDE4D, or KCNH2 protein levels. CONCLUSION: This study revealed the cellular atlas, immune microenvironment shifts, and potential target genes of WFC in treating CAG, demonstrating that WFC may delay CAG progression by downregulating MMP12 expression. Machine learning identified JUN as the candidate predictive gene. This research provides a preliminary approach for precision mechanism studies of traditional Chinese medicine formulations, encompassing ""single-cell atlas - multi-source data integration - machine learning prediction - experimental validation"".","[""Chen Haixin"", ""Zhou Chencong"", ""Wu Kaihan"", ""Ping Guoqing"", ""Jiang Chaohong"", ""Huang Xuan""]",Journal of ethnopharmacology,2026-09-09,10.1016/j.jep.2026.122381,"[""Journal Article""]","[""Gastritis, Atrophic"", ""Machine Learning"", ""Humans"", ""Drugs, Chinese Herbal"", ""Single-Cell Analysis"", ""Single-Cell Gene Expression Analysis"", ""Gastric Mucosa"", ""Cell Line""]", 42697496,https://pubmed.ncbi.nlm.nih.gov/42697496/,Breaking the autophagy-oxidative stress vicious cycle in Alzheimer's disease: Lactiflorin unlocks P62 dual phosphorylation via ULK1 targeting.,"ETHNOPHARMACOLOGICAL RELEVANCE: The dried root of Paeonia lactiflora Pall. has a long history of medicinal use in traditional Chinese medicine. Classical materia medica and traditional practice describe Paeonia-related medicinal materials for headache, dizziness, restlessness, and other neurological or behavioral manifestations, particularly in disorders traditionally associated with blood or yin deficiency and liver-yang hyperactivity. Modern pharmacological studies have further demonstrated neuroprotective and antioxidant effects of Paeonia lactiflora and its bioactive monoterpene glycosides. Lactiflorin (LAC) is a naturally occurring monoterpene glycoside reported as a constituent of Paeonia lactiflora and exhibits antioxidant and cytoprotective properties; however, its therapeutic potential and underlying mechanisms in Alzheimer's disease (AD) remain unclear. This study therefore investigated the protective effects of LAC against AD and explored its underlying mechanisms. AIM OF THE STUDY: This study investigated whether LAC could break the self-reinforcing vicious cycle between autophagic dysfunction and oxidative stress in AD by engaging ULK1 to promote P62 phosphorylation at two key residues. MATERIALS AND METHODS: APP/PS1 transgenic mice and Aβ1-42-treated HT22 cells served as in vivo and in vitro AD models, respectively, with cognitive performance assessed through a battery of behavioral tests. Integrated bioinformatics and machine-learning analyses were applied to map AD-related molecular networks and prioritize candidate targets. Aβ deposition, autophagic flux, and oxidative stress were evaluated by immunofluorescence, Western blotting, biochemical assays, and the mCherry-EGFP-LC3 tandem fluorescent reporter system. The interaction between LAC and ULK1 was subsequently evaluated by molecular docking, molecular dynamics simulation, thermal shift assay, DARTS, and SPR. RESULTS: LAC alleviated cognitive deficits in male APP/PS1 mice and reduced hippocampal Aβ deposition. Bioinformatics analysis subsequently suggested that LAC may regulate AD-related pathology mainly through autophagy- and oxidative stress-associated networks. Consistent with these predictions, LAC increased the LC3-II/LC3-I ratio and decreased P62 expression, while also alleviating redox imbalance and lipid peroxidation in brain tissue and HT22 cells, as evidenced by decreased malondialdehyde (MDA) levels and restored superoxide dismutase (SOD) and glutathione peroxidase (GSH-Px) activities. Mechanistically, LAC engaged ULK1 via a His24-dependent interaction, thereby enhancing P62 phosphorylation at Ser403/Ser351, activating the Keap1/Nrf2/HO-1 pathway, and restoring the Beclin-1/VPS34-associated autophagy-initiation machinery. CONCLUSIONS: LAC alleviated Aβ deposition, oxidative injury, and cognitive dysfunction by engaging ULK1, suggesting its potential as a promising natural compound for AD treatment.","[""Zhang Jie"", ""Chen Xinhua"", ""Hu Zuobin"", ""Zhang Jingjing"", ""Wu Tianchen"", ""Zhou Jing"", ""Song Bohui"", ""Xu Yun"", ""Yang Hui"", ""Zhu Xiaolei""]",Journal of ethnopharmacology,2026-09-04,10.1016/j.jep.2026.122367,"[""Journal Article""]","[""Animals"", ""Oxidative Stress"", ""Alzheimer Disease"", ""Phosphorylation"", ""Autophagy"", ""Mice, Transgenic"", ""Sequestosome-1 Protein"", ""Male"", ""Autophagy-Related Protein-1 Homolog"", ""Mice"", ""Amyloid beta-Peptides"", ""Cell Line"", ""Peptide Fragments"", ""Neuroprotective Agents"", ""Hippocampus""]", 42849109,https://pubmed.ncbi.nlm.nih.gov/42849109/,Tracing the roots of illicit Cannabis: A machine learning-assisted ATR-FTIR proof-of-concept study for geographical origin classification.,"Cannabis remains one of the most commonly trafficked and seized illicit drugs around the globe. The growing number of cannabis seizures has created an increasing demand for analytical methods able to determine the geographical origin of seized material rapidly and reliably. Existing approaches are often destructive, require extensive sample preparation, and have limited applicability to Indian cannabis. To address these limitations, the present study explored Fourier-transform infrared spectroscopy in attenuated total reflectance sampling mode (ATR-FTIR) combined with machine learning (ML) approaches for the geographical classification of cannabis samples collected from North India. A total of 200 female cannabis flower-top samples were collected from four distinct geographical regions and 1000 spectral observations were recorded using ATR-FTIR spectroscopy. The spectra obtained presented characteristic absorption bands related to cannabinoids, flavonoids, cellulose, hemicellulose, lignin, and other phytochemical constituents. Then, dataset was reduced using principal component analysis (PCA) and subsequently used to train four ML models. Among the models evaluated, the artificial neural network (ANN) exhibited the best classification performance during the training phase. Its reliability was further validated internally using test data and externally using unknown data. Further, PC3 was the most contributing component to the classification across all models and was associated with holocellulose, lignin, pectin, THCA, and THC. The proposed work has a strong potential for forensic investigations, especially for drug trafficking and identification of cannabis origin in illicit market.","[""Singh Mahan"", ""Sharma Akanksha"", ""Sharma Vishal""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-28,10.1016/j.saa.2026.128847,"[""Journal Article""]","[""Cannabis"", ""Spectroscopy, Fourier Transform Infrared"", ""Machine Learning"", ""Principal Component Analysis"", ""Neural Networks, Computer"", ""India"", ""Cannabinoids"", ""Geography"", ""Illicit Drugs"", ""Classification Algorithms"", ""Cellulose""]", 42837988,https://pubmed.ncbi.nlm.nih.gov/42837988/,Combining ATR-FTIR spectroscopy and DEGs with machine learning for necrophagous insect diapause duration estimation.,"Forensic entomology plays an important role in estimating the postmortem interval (PMI). However, during cold seasons, necrophagous insects may enter diapause, resulting in developmental arrest. Currently, effective methods for estimating diapause duration are limited, which may compromise the accuracy of PMI estimation based on insect evidence. Here we show that multi-dimensional data fusion of gene expression and infrared spectral features could improve the accuracy of diapause duration estimation. Using the forensically important flesh fly Sarcophaga peregrina as a model, we performed transcriptome sequencing across the diapause period and identified 14 core differentially expressed genes (DEGs) with pronounced temporal dynamics. After qPCR validation, 11 DEGs were retained. ATR-FTIR analysis revealed stage-dependent shifts in macromolecular absorption peak features. Source-level integration of the two data types, followed by partial least-squares dimensionality reduction and support vector regression, produced a model (test R2 = 0.922) that substantially outperformed either dataset alone. These findings suggest a potential molecular-spectroscopic strategy that may help address the long-standing challenge of PMI estimation in cold-season forensic cases, and offer a preliminary framework for exploring multi-dimensional data fusion in other complex biological processes.","[""Zhang Xiangyan"", ""Qu Hongke"", ""Chen Sile"", ""Wang Hao"", ""Xiao Yuyan"", ""Zhou Ziqi"", ""Zhang Changquan"", ""Guo Yadong""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-30,10.1016/j.saa.2026.128877,"[""Journal Article""]","[""Animals"", ""Spectroscopy, Fourier Transform Infrared"", ""Forensic Entomology"", ""Machine Learning"", ""Diapause, Insect"", ""Diptera"", ""Gene Expression Profiling"", ""Transcriptome""]", 42837984,https://pubmed.ncbi.nlm.nih.gov/42837984/,A robust ensemble chemometric strategy for near-infrared determination of ash content in fishmeal.,"Near-infrared (NIR) spectroscopy combined with machine learning algorithms has been widely used for rapid and non-destructive quantitative analysis, but accurate quantitative modeling remains challenging due to the high dimensionality, collinearity, and redundancy of spectral variables, as well as the limited size of calibration datasets. In this study, a customized ensemble framework termed RFXGB-DA was developed for predicting fishmeal ash content. The proposed method combines the bootstrap sampling and random feature selection mechanisms of Random Forest (RF) with the nonlinear regression capability of eXtreme Gradient Boosting (XGBoost), and incorporates a random mixed data augmentation strategy to simulate realistic spectral perturbations. This framework aims to reduce the influence of redundant variables, enhance model diversity, achieve a better balance between bias and variance, and improve robustness against sample heterogeneity and measurement variability. The performance of RFXGB-DA was evaluated on a fishmeal NIR dataset and compared with Partial Least Squares (PLS), RF, XGBoost, and RFXGB. The results showed that RFXGB-DA achieved the best predictive performance, with RT2 = 0.9375 ± 0.0146, RMSET = 0.9317 ± 0.1067 and RPD = 4.0766 ± 0.4527. In addition, ablation analysis demonstrated that both the ensemble mechanism and data augmentation played significant roles in improving calibration performance. SHAP analysis further indicated that the model captured chemically meaningful spectral regions associated with indirect responses from the organic matrix and water related to minerals. Overall, the proposed method offers a robust and interpretable chemometric strategy for the rapid determination of ash content in fishmeal using NIR spectroscopy.","[""Li Linghui"", ""Yang Lei"", ""Li Wuchen"", ""Zhang Ruimin"", ""Li Chichi"", ""Chen Huazhou""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-25,10.1016/j.saa.2026.128850,"[""Journal Article""]",, 42833091,https://pubmed.ncbi.nlm.nih.gov/42833091/,Error-diagnosis-guided cross-indicator adaptive calibration for near-infrared spectroscopic multi-property prediction of coal quality.,"Near-infrared spectroscopy (NIRS) coal-quality datasets often contain many repeated spectra acquired from the same coal samples. Ignoring the coal-sample group structure and indicator-dependent error patterns may compromise the reliability of generalization assessment and calibration stability. To address this issue, this study proposes an error-diagnosis-guided cross-indicator adaptive calibration (ED-CIAC) method for NIRS-based multi-property prediction of coal quality. ED-CIAC performs coal-sample-group-level modeling by robustly aggregating repeated spectra and integrating complementary spectral representations to obtain property-specific base predictions. The remaining prediction errors are characterized across properties and value ranges to identify systematic residual structures. Cross-indicator features constructed from the multi-property base predictions are subsequently introduced as complementary calibration information. Candidate calibrators with direct and residual learning modes are evaluated within each training fold, and the calibrator, learning mode, and correction strength are jointly selected for each property according to internal validation performance. Experiments were conducted on 26 batches of NIRS coal-quality data, including 122,983 repeated spectra from 538 independent coal-sample groups, using coal-sample-group-level fivefold cross-validation. The proposed method achieved R2 values of 0.8717 ± 0.0227, 0.9077 ± 0.0120, 0.8112 ± 0.0460, and 0.9000 ± 0.0157 for total moisture (Mt), dry-basis ash (Ad), dry-basis volatile matter (Vd), and net calorific value as received (Qnet,ar), respectively, outperforming conventional chemometric, machine learning, and literature-based comparison methods. These results demonstrate that group-level multi-view base prediction, residual-guided property-specific calibration, and cross-indicator information utilization can improve the reliability of NIRS-based multi-property coal-quality prediction.","[""Weng Zhipeng"", ""Li Jiangyun"", ""Yuan Li""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-28,10.1016/j.saa.2026.128894,"[""Journal Article""]",, 42833085,https://pubmed.ncbi.nlm.nih.gov/42833085/,Kubelka-Munk theory enhances hyperspectral mineral identification for rock heritage.,"Accurate identification of minerals is the fundamental task in the protection of rock cultural relics, directly determining the scientific validity and effectiveness of research on weathering mechanisms, disease prediction, and the development of protective materials. However, remote sensing technology is constrained by spatial and observational conditions, making fine in‑situ surveys difficult to achieve. In contrast, proximal hyperspectral imaging overcomes these limitations, providing a reliable technical foundation and data source for in‑situ mineral detection. In this study, a hyperspectral database of common minerals spanning the 900-1700 nm band was established. The diagnostic spectral features of minerals were enhanced through the integration of Kubelka-Munk theory, while efficient mineral identification was achieved by applying machine learning models. Experimental results indicate that diagnostic spectral characteristics are most pronounced within the 1360-1490 nm band. Performance evaluation shows that average efficiency increased by 12.55% and 28.08%, mean F1 scores improved by 12% and 6.14%, mean Kappa values increased by 10.86% and 4.43%, and the F1score of SVM improved by 30%. In addition, spectral detection was conducted on sandstone samples from the Dazu stone carvings, and the experimental results confirmed the accuracy and reliability of the proposed method.","[""Peng Lixin"", ""Yang Haiqing"", ""Zhou Xiaoyu"", ""Li Xingyue"", ""Ju Anhua"", ""Jiang Zhonghua"", ""Yang Yang""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-30,10.1016/j.saa.2026.128903,"[""Journal Article""]",, 42830003,https://pubmed.ncbi.nlm.nih.gov/42830003/,Machine learning algorithms empower traditional Chinese medicine: Breakthrough approaches from chemical component analysis to pharmacological mechanism research.,"Traditional Chinese medicine studies are constrained by complex chemical components and multi-target biological interactions, making conventional analytical methods inefficient for mechanism exploration. As a data-driven technique, machine learning has been widely adopted to tackle diverse complex challenges in TCM research. Nevertheless, most existing reviews merely summarize individual algorithms or scattered application cases, lacking systematic task-based classification and unified operational standards. Based on literature published from 2022 to 2026, this review systematically classifies machine learning applications in TCM into four core research tasks. A task-oriented framework is constructed to match different algorithms with appropriate TCM research scenarios. This paper further establishes a complete standardized workflow covering algorithm selection, model construction and model evaluation to standardize ML applications in TCM research. This review also summarizes the applications of machine learning in TCM quality control, bioactive compound identification, processing analysis, and mechanism elucidation. Furthermore, major bottlenecks in current ML-based TCM studies are analyzed, and feasible optimization strategies are proposed. This review aims to provide clear and practical guidelines for the standardized application of machine learning in future TCM research.","[""Han Zheyue"", ""Zhao Fuxia"", ""Xu Haonan"", ""Zhang Zhenkai"", ""Wu Jiatong"", ""Zhang Yiqiang"", ""Pan Juan"", ""Li Mengmeng"", ""Hao Zhichao"", ""Guan Wei"", ""Jiang Peng"", ""Luo Yumeng"", ""Ma Xiaochi"", ""Kuang Haixue"", ""Liu Yan""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-27,10.1016/j.saa.2026.128859,"[""Journal Article"", ""Review""]","[""Medicine, Chinese Traditional"", ""Machine Learning"", ""Algorithms"", ""Drugs, Chinese Herbal"", ""Humans"", ""Soft Computing""]", 42828937,https://pubmed.ncbi.nlm.nih.gov/42828937/,Chemometrics-assisted FTIR and machine learning for biochemical profiling of human skeletal muscles in relation to type 2 diabetes.,"Skeletal muscle is the major site of insulin-stimulated glucose uptake, making it highly susceptible to metabolic disorders. This study investigates Type 2 diabetes mellitus-associated alterations in the macromolecular composition of functionally distinct human skeletal muscles. These modifications are crucial to understanding the molecular basis of diabetic myopathies but remain underexplored. Five skeletal muscles from 80 individuals (40 controls and 40 individuals with Type 2 diabetes) were analysed. To overcome the limitations of traditional histochemical assays, we employed infrared spectroscopy combined with a chemometric decomposition approach to resolve spectra into five spectral components, each dominated by a specific macromolecular group. The corresponding component weights quantified and compared relative macromolecular proportions between the diabetic and control groups for different muscle types. Compared with controls, muscles from individuals with diabetes showed a consistent pattern of higher lipid levels and lower glycogen levels, with a statistically significant increase in lipids within the diaphragm. This biochemical profiling was enriched by building machine learning classifiers, which achieved low-to-modest but statistically significant AUC performance of up to 72% (p < 0.05) in discriminating diabetic from healthy muscle. Our work demonstrates an efficient methodology for the simultaneous screening of various macromolecular constituents using spectral data obtained in a single experiment. It provides medically relevant insights into the biological patterns of diabetes-associated skeletal muscle remodelling to guide future research. However, an independent validation should be performed to overcome this study limitation and draw definitive conclusions.","[""Abdelmonaem Mohamed Elwy Abdelhamed"", ""Ugwoke Chiedozie Kenneth"", ""Alibegović Armin"", ""Cvetko Erika"", ""Umek Nejc"", ""Grdadolnik Jože"", ""Zupančič Barbara""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-28,10.1016/j.saa.2026.128893,"[""Journal Article""]","[""Humans"", ""Diabetes Mellitus, Type 2"", ""Muscle, Skeletal"", ""Machine Learning"", ""Spectroscopy, Fourier Transform Infrared"", ""Female"", ""Male"", ""Middle Aged"", ""Adult"", ""Case-Control Studies"", ""Aged"", ""Lipids""]", 42810038,https://pubmed.ncbi.nlm.nih.gov/42810038/,Multimodal deep learning-driven Raman spectroscopy for rapid identification of closely related pathogenic Bacillus species.,"Rapid and fine-grained identification of closely related pathogenic Bacillus species is a critical challenge in optical biosensing and biomolecular spectroscopy, hindered by their highly similar biochemical compositions and severe spectral overlap. Existing single-modality spectroscopic analysis and computational frameworks frequently fail to resolve these subtle spectral shifts, particularly when confronted with instrumental noise, background interference, and limited training data. To overcome these spectroscopic bottlenecks, this study proposes the Spectral Continuous Wavelet Transform Transformer Network (Spec-CWT-TransNet), a novel multimodal deep learning framework designed for the precise identification of closely related microbial strains. By transforming 1D Raman spectra into 2D wavenumber-scale scalograms utilizing the Continuous Wavelet Transform (CWT), the proposed method synergizes the robust local topological texture extraction of Convolutional Neural Networks (CNNs) with the global dependency modeling capabilities of Transformers. Furthermore, a Fisher Score-based scale optimization strategy is introduced to adaptively select discriminative frequency bands, effectively suppressing redundant instrumental and background noise. Experimental results demonstrate that Spec-CWT-TransNet achieves a superior overall identification accuracy of 99.38% on four Bacillus species. Notably, under limited data conditions (using only 40% of training data), the framework maintains a robust accuracy of 92.50%, quantifiably surpassing the performance of traditional 1D baselines trained on the full dataset. Moreover, under severe spectral interference, the model retains 90.63% accuracy, outperforming unimodal baseline models by margins exceeding 22.5%. Overall, this multimodal Raman data analysis methodology demonstrates great potential for rapid, robust, and reproducible spectral discrimination in non-destructive optical sensing and bioanalytical applications.","[""Sun Jiazheng"", ""Liu Haodong"", ""Yang Kaier"", ""Ma Xiangyun"", ""Li Qifeng""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-27,10.1016/j.saa.2026.128883,"[""Journal Article""]","[""Spectrum Analysis, Raman"", ""Bacillus"", ""Deep Learning"", ""Convolutional Neural Networks"", ""Wavelet Analysis""]", 42810037,https://pubmed.ncbi.nlm.nih.gov/42810037/,Raman spectroscopy and multivariate analysis for quantifying uranium(VI) under uranyl nitrate crystallization-relevant conditions.,"A fiber-optic Raman spectroscopy approach was developed for solution-phase, in situ quantification of U(VI) under uranyl nitrate crystallization-relevant conditions, where temperature fluctuations and chemically complex matrices can compromise calibration performance. A multivariate regression framework is presented that combines Raman spectroscopy, design of experiments, and chemometrics to predict U(VI) concentration in HNO3 in pre- and post-crystallization regimes. A D-optimal design was used to efficiently select calibration concentrations spanning the relevant composition space. Principal component analysis of Raman spectra acquired over a range of concentrations and temperatures revealed systematic trends, including a temperature-dependent redshift in the ν1 symmetric U(VI) stretch at high U(VI) concentrations near the solubility limit. Partial least squares regression (PLSR) and support vector regression (SVR) models were developed and evaluated from 20 to 60 °C in the presence of surrogate fission products (Cs, Sr, Ru, Nd, Re, Zr, and Mo) and varying HNO3 levels. SVR outperformed PLSR for U(VI) and HNO3 quantification by delivering strong accuracy despite temperature- and matrix-induced spectral variation. These results highlight the utility of Raman spectroscopy coupled with multivariate calibration for solution-phase quantification under uranyl nitrate crystallization-relevant conditions and provide a basis for future monitoring applications in used nuclear fuel recycling and other crystallization-focused process analytics settings.","[""Russell David V"", ""Sadergaski Luke R"", ""Einkauf Jeffrey D"", ""Pyles Jennifer M"", ""Andrews Hunter B"", ""Delmau Laetitia H"", ""Burns Jonathan D""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-11,10.1016/j.saa.2026.128778,"[""Journal Article""]",, 42810032,https://pubmed.ncbi.nlm.nih.gov/42810032/,"Multi-source fusion of NIR and UV spectroscopy with refractometry for pooled prediction of total flavonoids, total polyphenols and solids content in herbal water extracts.","Rapid quantification of multiple quality indicators in compositionally distinct herbal formulas calls for predictive approaches that can accommodate pooled samples within a shared modeling framework. To address this limitation, the present study integrates multi-source information from near-infrared (NIR) spectroscopy, ultraviolet (UV) spectroscopy, and refractive index (RI) measurements to develop one pooled model for each of three indicators: total flavonoids, total polyphenols and total solids content in aqueous extracts of six TCM formulas. Six TCM formulas with markedly distinct herbal compositions were selected, and 216 samples were prepared through a gradient extraction protocol. Under a unified SGS + FD pretreatment, three fusion architectures, five feature selection algorithms, and 16 machine learning models were systematically compared. Low-level fusion (LowFusion) performed best for flavonoids and polyphenols before feature selection, whereas mid-level fusion performed slightly better for solids content; the optimal feature selection algorithm was analyte-dependent, and the selected wavelength regions were consistent with plausible electronic-transition and vibrational assignments. Validation using 18 independent samples from new batches of the same six formulas showed strong predictive performance for flavonoids and solids content. The polyphenol model displayed systematic underestimation in the high-concentration range. These results support pooled multi-formula modeling and independent-batch prediction within the investigated formula set, providing a basis for rapid multi-indicator quality assessment without developing a separate model for each formula.","[""Gao Wufeng"", ""Cheng Long"", ""Dong Yuxuan"", ""Tao Junzhang"", ""Wu Yingjun"", ""Fang Can"", ""Xian Jiechen"", ""Lin Xiao"", ""Shen Lan"", ""Wang Youjie"", ""Hong Yanlong""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-26,10.1016/j.saa.2026.128858,"[""Journal Article""]","[""Flavonoids"", ""Polyphenols"", ""Spectroscopy, Near-Infrared"", ""Spectrophotometry, Ultraviolet"", ""Refractometry"", ""Drugs, Chinese Herbal"", ""Plant Extracts"", ""Water"", ""Algorithms""]", 42800135,https://pubmed.ncbi.nlm.nih.gov/42800135/,Application of a semi-supervised learning framework combining principal component constraint enhancement in near-infrared spectroscopy of Rhizome Coptidis.,"Rhizome Coptidis is a popular traditional Chinese medicinal herb, and analyzing its composition is crucial for evaluating its quality. However, existing analytical methods can produce significantly different results due to various practical factors, so there is an urgent need to develop more suitable analytical methods. Near-infrared spectroscopy (NIR) is widely used in the analysis of traditional Chinese medicine due to its rapid, non-destructive, and pollution-free characteristics. It offers numerous possibilities for the detection of Rhizome Coptidis. However, this technology often encounters challenges such as small sample sizes, limited annotated data, and redundant information. More advanced modeling methods are therefore required to address these issues. Based on this, this paper proposes a fusion framework for the quantitative analysis of Rhizome Coptidis: Spectral Augmented Principal Component Analysis-Semi-Supervised Learning-Convolutional Neural Network-Twin-Kernel Support Vector Regression (SAPCA-SSLCNN-TKSVR). This framework comprises three modules: Spectral Augmented Principal Component Analysis (SAPCA) for data augmentation, Semi-Supervised Learning Convolutional Neural Networks (SSLCNN) for feature extraction, and Twin Kernel Support Vector Regression (TKSVR) for data prediction. The experimental results demonstrate that this method is well-suited to the quantitative analysis of Rhizome Coptidis, particularly in scenarios involving small sample sizes, limited labeled data, and redundant information. It effectively improves modeling performance for high-dimensional, highly linear data and offers a viable approach to the field of near-infrared spectroscopy.","[""An Nan"", ""Zhang Jingyu"", ""Chen Huazhou""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-23,10.1016/j.saa.2026.128843,"[""Journal Article""]","[""Principal Component Analysis"", ""Spectroscopy, Near-Infrared"", ""Rhizome"", ""Drugs, Chinese Herbal"", ""Supervised Machine Learning"", ""Convolutional Neural Networks"", ""Support Vector Machine""]", 42800132,https://pubmed.ncbi.nlm.nih.gov/42800132/,Understanding calibration robustness under extreme spectral multicollinearity: A systematic evaluation of high-dimensional NIR spectroscopy.,"High-dimensional near-infrared (NIR) spectroscopy datasets are characterized by strong wavelength correlation, limited sample sizes, and structured experimental variability, creating significant challenges for multivariate calibration. Under these conditions, extreme multicollinearity can lead to unstable regression coefficients, inflated estimator variance, and reduced model robustness. Although numerous calibration algorithms have been proposed, comparatively little attention has been devoted to understanding how multicollinearity influences calibration behaviour across different modelling paradigms. This study systematically investigates the performance of classical chemometric, regularized linear, dimensionality reduction, wavelength selection, and nonlinear machine learning approaches using a publicly available glucose NIR spectroscopy dataset comprising 4200 wavelength variables and structured experimental perturbations. Partial Least Squares (PLS), Ridge Regression, Elastic Net, Principal Component Analysis (PCA), Genetic Algorithm-based PLS (GA-PLS), Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest, Gradient Boosting, and Gaussian Process Regression were evaluated using repeated cross-validation and independent validation. Model assessment was complemented by multicollinearity diagnostics, residual analysis, and wavelength-selection stability analysis to investigate the mechanisms governing calibration robustness. SNV-PLS achieved the best overall predictive performance on the independent validation dataset (RMSE = 5.75 mM; R2 = 0.864), followed by GA-PLS (RMSE = 5.82 mM; R2 = 0.861) and SNV-Ridge (RMSE = 5.84 mM; R2 = 0.860), while SNV-Elastic Net achieved comparable performance (RMSE = 5.93 mM; R2 = 0.856). GA-PLS reduced the spectral dimensionality from 4200 to 50 wavelengths; however, repeated GA runs showed only 0-2% exact wavelength overlap, increasing to 17-29% within ±5 nm and 30-47% within ±10 nm, indicating regional rather than exact-wavelength stability. Variance Inflation Factor analysis revealed extreme predictor collinearity, while residual and domain-coverage analysis showed that the major validation deterioration was localized at 50 mM glucose, where the six validation samples were systematically underpredicted and originated from an experimental condition absent from the calibration set. The findings suggest that effective management of covariance structure and estimator variance may be more influential than increasing model complexity for the investigated high-dimensional NIR calibration problem. These results provide practical guidance for selecting robust and interpretable calibration strategies for high-dimensional spectroscopic applications affected by severe multicollinearity.","[""Naqvi Ila"", ""Singh Rajshree"", ""Jain Pallavi"", ""Mehdi Yusuf""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-24,10.1016/j.saa.2026.128854,"[""Journal Article""]",, 42790015,https://pubmed.ncbi.nlm.nih.gov/42790015/,An ATR-FTIR spectroscopy strategy coupled with ensemble learning for skin contusion age estimation under a 24 h Postmortem interval.,"Accurate estimation of wound age is a critical issue in forensic practice. However, injured tissues in real-world cases are often affected by postmortem changes, which may reduce the reliability of conventional laboratory methods. This study established a skin contusion model in SD rats, collected skin tissue samples under different wound ages (0-48 h) and postmortem intervals (0 h and 24 h), utilized ATR-FTIR to obtain molecular spectral information, and combined it with 10 machine learning algorithms to construct predictive models. The results demonstrated that 24 h postmortem changes affected the spectral characteristics of skin contusion tissues, resulting in decreased predictive accuracy of wound age estimation models. To reduce the interference caused by postmortem alterations, a prediction framework integrating genetic algorithm (GA)-based feature selection and ensemble learning strategies was further developed. The Average ensemble model composed of GA-LightGBM and GA-SVR exhibited satisfactory predictive performance, achieving CV-R2 = 0.89 and CV-RMSE = 5.11 h in five-fold cross-validation, and Test-R2 = 0.85 and Test-RMSE = 5.97 h on the independent test set. SHAP analysis revealed that protein-related features gradually decreased with increasing wound age, whereas lipid- and phosphate-related features gradually increased. Ultimately, this study developed a predictive model capable of mitigating the effects of 24 h postmortem changes and enabling reliable estimation of skin contusion wound age.","[""Wei Bi"", ""Qian Xuehong"", ""Chen Hongbing"", ""Wang Yusen"", ""Xiong Hongli"", ""Guo Jiang"", ""Xiao Hao"", ""Yu Kai"", ""Li Jianbo""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-19,10.1016/j.saa.2026.128808,"[""Journal Article""]","[""Animals"", ""Contusions"", ""Rats, Sprague-Dawley"", ""Skin"", ""Spectroscopy, Fourier Transform Infrared"", ""Rats"", ""Postmortem Changes"", ""Ensemble Learning"", ""Machine Learning"", ""Male"", ""Time Factors""]", 42785017,https://pubmed.ncbi.nlm.nih.gov/42785017/,Hyperspectral imaging combined with data fusion for non-destructive prediction of soluble solids content and water-soluble sugars in strawberries during storage after static-pressure bruising.,"Soluble solid content (SSC) and water-soluble sugar contents are key indicators for evaluating strawberry quality. Static-Pressure Bruising can alter these components, thereby affecting fruit texture and overall eating quality. This study aimed to investigate the feasibility of predicting SSC as well as fructose and glucose contents in strawberries subjected to Static-Pressure Bruising followed by different storage durations using two hyperspectral imaging techniques (Vis-NIR and NIR) combined with low-level and mid-level data fusion strategies (LLF and MLF). After applying five spectral preprocessing methods and three characteristic wavelength selection approaches, two conventional machine-learning models and four deep-learning models were developed. The results showed that the data fusion strategies outperformed single-spectrum models for all prediction targets. The optimal SSC prediction model was LLF combined with IVSO-LSSVM, achieving an Rp of 0.8930, an RMSEP of 0.4982°Brix, and an RPD of 2.2312. For fructose prediction, the best model was MLF combined with IVSO-TCN-BiGRU-Self-Attention, with an Rp of 0.7691, an RMSEP of 3.8737 mg/g, and an RPD of 1.5707. For glucose prediction, the optimal model was LLF combined with IVSO-LSSVM, yielding an Rp of 0.7435, an RMSEP of 3.0028 mg/g, and an RPD of 1.5014. Based on the optimal models, visualization maps of SSC and water-soluble sugar distributions in strawberries were generated. These results demonstrate the feasibility of hyperspectral imaging combined with data fusion strategies for monitoring quality variations in strawberries subjected to static pressure bruising, although further optimization is needed to achieve accurate quantitative performance for individual sugars.","[""Liu Xin"", ""Chen Meng"", ""Hao Zhengxiong"", ""Zhang Yufei"", ""Zhao Zhilei"", ""Zhao Xin""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-17,10.1016/j.saa.2026.128807,"[""Journal Article""]","[""Fragaria"", ""Hyperspectral Imaging"", ""Water"", ""Solubility"", ""Pressure"", ""Fruit"", ""Food Storage"", ""Sugars"", ""Glucose"", ""Fructose"", ""Spectroscopy, Near-Infrared""]", 42785013,https://pubmed.ncbi.nlm.nih.gov/42785013/,A lightweight spectral qualitative analysis method based on Res-Inception.,"Infrared spectroscopy is widely used for qualitative classification tasks in pharmaceutical, food, agricultural product and other fields, owing to its inherent technical advantages of rapid detection, non-destructiveness, reagent-free and pollution-free operation, as well as reliable in-situ detection capability. However, traditional qualitative spectral analysis methods rely heavily on manual feature extraction and shallow machine learning models, making it difficult to fully capture the nonlinear features embedded in high-dimensional spectral data, with their generalization performance severely constrained by prior domain knowledge and dataset characteristics. Although existing deep learning methods have improved classification accuracy, most models are custom-developed for specific tasks in a single domain, resulting in insufficient multi-domain generalizability, while their large parameter count and computational complexity restrict edge deployment on portable spectral instruments and engineering-scale promotion. To address the above limitations, this paper proposes DSRINet-1D, a lightweight one-dimensional (1D) spectral classification network integrating depthwise separable convolution, Inception multi-scale structure, and nonlinear residual connection, which drastically reduces model computational complexity and parameter count while ensuring classification accuracy and universality. A Feature Screening and Weight Assignment (FSWA) attention mechanism is further designed, which dynamically optimizes the weight allocation of multi-scale features, significantly enhancing the model's feature representation capability with only a slight parameter increase, thus achieving an efficient balance between classification performance and lightweight design. Experimental results on three cross-domain public datasets, namely Tablets, fruit juice, and meat, demonstrate that the FSWA-embedded DSRINet-1D achieves classification accuracies of 99.355%, 98.474% and 99.726% respectively, with comprehensive performance significantly superior to traditional machine learning methods and mainstream deep learning models. Compared with the baseline Res-Inception model, the basic DSRINet-1D reduces parameter count by 42.4% and floating-point operations (FLOPs) by 94.4%, exhibiting a remarkable lightweight advantage. This model requires no complex manual feature selection or preprocessing procedures, boasts strong generalization capability, and is easy to deploy for rapid field applications. It also provides a novel research insight for spectral classification tasks with different input forms, and explores a two-dimensional spectral modeling method based on dual-band transformation.","[""Song Gaokui"", ""Wang Nan"", ""Yang Xinwu""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-17,10.1016/j.saa.2026.128811,"[""Journal Article""]",, 42785011,https://pubmed.ncbi.nlm.nih.gov/42785011/,Chemical fingerprinting of rapeseed varieties: A BiLSTM-based fusion strategy for near-infrared and fluorescence spectra.,"To achieve rapid and accurate identification of rapeseed (Brassica napus) varieties, this study proposes a classification strategy based on multi-source spectral data fusion and deep learning algorithms. Near-infrared (NIR, 4000-11,550 cm-1) and fluorescence spectra (FS, 400-850 nm) of seven rapeseed varieties were collected, and the performances of five classification models-support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN), convolutional neural network (CNN), and bidirectional long short-term memory (BiLSTM)-were systematically compared. Two fusion strategies-low-level data fusion and feature-level fusion-were employed in combination with preprocessing methods such as multiplicative scatter correction (MSC) to optimize spectral quality and enhance model discriminative capability. The results demonstrated that multi-source spectral data fusion improved classification accuracy, with CNN and BiLSTM models showing the best performance, both achieving an accuracy of 98.41% after MSC preprocessing. Furthermore, the successive projections algorithm (SPA) was applied for characteristic wavelength selection, and when integrated with the BiLSTM-based feature-level fusion framework, the highest classification accuracy of 99.21% was achieved. This study supports the feasibility and potential advantages of NIR-FS fusion for rapeseed variety identification, providing an efficient and rapid spectroscopic approach for agricultural product quality assurance and cultivar authenticity verification.","[""Su Wenlong"", ""Suo Limin"", ""Qian Lili"", ""Liu Hailong"", ""Sun Liyuan""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-21,10.1016/j.saa.2026.128827,"[""Journal Article""]","[""Spectroscopy, Near-Infrared"", ""Spectrometry, Fluorescence"", ""Support Vector Machine"", ""Brassica napus"", ""Convolutional Neural Networks"", ""Classification Algorithms"", ""Long Short Term Memory"", ""Random Forest"", ""Algorithms""]", 42785009,https://pubmed.ncbi.nlm.nih.gov/42785009/,Plasma ATR-FTIR spectroscopy combined with machine learning for nasopharyngeal carcinoma classification.,"BACKGROUND: Nasopharyngeal carcinoma (NPC) is often diagnosed at advanced stages, creating a need for minimally invasive approaches to support detection. This study evaluated plasma attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy, combined with machine learning, to distinguish NPC from clinically healthy individuals. METHODS: Plasma samples from 51 histologically confirmed NPC cases and 51 age- and sex-matched clinically healthy controls were analyzed across 4000-600 cm-1. Differences at 21 spectral peaks were assessed using the Mann-Whitney U test. Seven machine-learning algorithms were evaluated using the full spectrum, fingerprint region, selected peaks, and low-rank spectral representations with repeated cross-validation and exploratory age- and sex-stratified analyses. RESULTS: Thirteen peaks showed significant between-group differences in median absorbance (p < 0.01). Despite substantial overlap in the original spectral distributions, model performance in the complete cohort increased when selected peaks and low-rank spectral representations were used: the best case AUCs were 0.6303 ± 0.0384 for the full spectrum, 0.6330 ± 0.0333 for the fingerprint region, 0.7066 ± 0.0452 for selected peaks, and 0.7404 ± 0.0422 for the low-rank representation. The neural network model achieved the best overall performance using low-rank features, with an accuracy (ACC) of 0.7115 ± 0.0456. Exploratory subgroup estimates varied across age- and sex-defined subsets. However, due to limited subgroup sizes and variability in the performance estimates, interpretation and generalizability was cautioned. CONCLUSION: Plasma ATR-FTIR spectroscopy combined with machine learning showed moderate discrimination between NPC and clinically healthy controls, with the highest performance obtained using low-rank spectral features and a feedforward neural network. Larger independent cohorts are needed for validation.","[""Tomas Rock Christian"", ""Suzuki Yohsuke"", ""Magno Gerard Mathew"", ""Lee Michael"", ""Millan Javier Alfonso"", ""Nepomuceno Keziah"", ""Cobarrubias Mariella Cielo"", ""Mejia Michael Benedict"", ""Sy-Ortin Teresa"", ""Albano Pia Marie""]","Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy",2026-09-21,10.1016/j.saa.2026.128835,"[""Journal Article""]","[""Humans"", ""Spectroscopy, Fourier Transform Infrared"", ""Machine Learning"", ""Female"", ""Male"", ""Nasopharyngeal Carcinoma"", ""Middle Aged"", ""Nasopharyngeal Neoplasms"", ""Adult"", ""Neural Networks, Computer"", ""Case-Control Studies"", ""Aged""]", 42815691,https://pubmed.ncbi.nlm.nih.gov/42815691/,Ultrasonic vocalizations as markers of anxiety and depression in rodent models of neurotrauma.,"Spinal cord injury (SCI) and traumatic brain injury (TBI) are severe events that cause lasting sensorimotor and autonomic impairments. Both injuries often also lead to psychological and somatosensory dysfunctions such as anxiety, depression, or pain. Preclinical assessment of anxiety and depression in rodents often relies on locomotion-based behavioural tests confounded by motor deficits and pain-related changes. Because rodents cannot self-report sensory or emotional experiences, behavioural outcomes are limited to observable proxies of anxiety and depression. Ultrasonic vocalizations (USVs) offer a complementary, non-locomotor tool that may capture anxiety- and depression-relevant context not detected by behavioural tests. Despite extensive USV research in stress, reward, and social behaviour, applications to neurotrauma remain underexplored, with no studies directly examining USVs in relation to anxiety- or depression-like outcomes after SCI. Even so, evidence from related fields and the limited neurotrauma literature suggests that USVs warrant further exploration as affective readouts after TBI and particularly SCI. USVs may complement behavioural testing by providing socioaffective context without relying on locomotor performance, potentially helping distinguish post-injury anxiety- and depression-like phenotypes from pain-related and other injury-associated changes. By integrating behavioural, neural, and pharmacological evidence, we highlight how USV features may reflect affective dimensions relevant to post-injury anxiety- and depression-like states and outline methodological considerations for validity and specificity. Machine-learning approaches further enable high-throughput, objective detection of subtle affective shifts. With development of collaborative datasets, these methods could reveal consistent affective vocal patterns and lay groundwork for a rudimentary 'lexicon' of rodent affect. Integrating USV analysis into neurotrauma and CNS disease research may refine preclinical models of affective dysfunction and improve triangulated evaluation of treatments targeting chronic mental health outcomes.","[""Scatterty Kale R"", ""Taylor Anna M W"", ""Fouad Karim""]",Behavioural brain research,2026-09-30,10.1016/j.bbr.2026.116510,"[""Journal Article"", ""Review""]","[""Animals"", ""Vocalization, Animal"", ""Anxiety"", ""Disease Models, Animal"", ""Depression"", ""Spinal Cord Injuries"", ""Brain Injuries, Traumatic""]", 43056665,https://pubmed.ncbi.nlm.nih.gov/43056665/,Large Language Models in Clinical Neurology: A Scoping Review.,"PURPOSE OF REVIEW: Large language models (LLMs) are increasingly being evaluated for clinical neurology applications, yet their real-world utility, safety, and deployment readiness remain unclear. This review maps the literature on LLMs applied to text-based and text-derived neurology tasks, characterizing current use cases, evaluation strategies, performance metrics, and reported safety concerns. We identify areas with relatively robust evidence, highlight important evidence gaps, and outline key methodological and clinical barriers that must be addressed before responsible clinical adoption. RECENT FINDINGS: Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR), we searched 4 databases (January 2022 to February 2026) and included 32 studies (2023-2026) spanning 7 neurology subspecialties; 9 were simulation or feasibility studies, 17 analyzed retrospective clinical data, and 6 reported prospective evaluation. Proprietary models predominated; 5 studies used retrieval-augmented generation and 1 used a multiagent framework. Performance was highest for constrained tasks, including binary diagnostic classification (area under the curve 0.75-0.94), information extraction (F1 0.89-0.90), patient-education question answering (accuracy 68%-97%), and ischemic stroke thrombectomy decision support (area under the curve [AUC] 0.92) and lowest for open-ended case-based reasoning (accuracy 42%-54%). Recurrent safety signals included hallucinated and fabricated citations, poor calibration with overconfidence (for example, 100% sensitivity with 27% specificity for epilepsy diagnosis), and a systematic bias toward intervention. On adapted Quality Assessment of Diagnostic Accuracy Studies Using Artificial Intelligence (QUADAS-AI) charting, overall risk of bias was low in 1 study (3%), moderate in 16 (50%), and high in 15 (47%), driven chiefly by absent external validation (31/32, 97%). Benchmarked against high-reliability and 6 Sigma standards, reported error rates remain orders of magnitude too high for autonomous use. SUMMARY: Evidence for LLMs in neurology is early, heterogeneous, and limited by scarce external and prospective validation. Current data support only narrow, clinician-supervised uses such as documentation and education, while cautioning against autonomous diagnostic or therapeutic use. Rigorous, prospective, externally validated studies with standardized reporting and explicit safety and ethical evaluation are needed before broader adoption. REGISTRATION: PROSPERO CRD420251082465.","[""Gorenshtein Alon"", ""Shihada Kamel"", ""Omar Mahmud"", ""Barash Yiftach"", ""Nadkarni Girish N"", ""Klang Eyal""]",Neurology. Clinical practice,2026-11-19,10.1212/CPJ.0000000000200687,"[""Scoping Review"", ""Journal Article"", ""Review""]","[""Large Language Models"", ""Humans"", ""Neurology""]", 43045438,https://pubmed.ncbi.nlm.nih.gov/43045438/,Artificial Intelligence and Sleep.,"This article explores how generative artificial intelligence (AI) and large language models are impacting the practice of sleep medicine from data analysis to patient interaction. The various AI advances in the field should be recognized as tools that have the potential to enhance care through increasing diagnostic expertise and reducing administrative burden. However, application in clinical settings require clinician familiarity and oversight to mitigate risks. As AI usage has become so ubiquitous and accessible, clinicians must navigate the blurring lines between patient-generated and clinical data, while preparing for a future where usage of AI may become a medicolegal necessity.","[""Schneider Logan Douglas"", ""Hernandez John"", ""Heneghan Conor""]",Neurologic clinics,2026-09-03,10.1016/j.ncl.2026.07.015,"[""Journal Article"", ""Review""]","[""Humans"", ""Artificial Intelligence"", ""Sleep"", ""Sleep Wake Disorders"", ""Generative Artificial Intelligence"", ""Large Language Models""]", 43045419,https://pubmed.ncbi.nlm.nih.gov/43045419/,The Two AIs: Applying Artificial Intelligence to Personalize Allergy and Immunology Care.,"Artificial intelligence (AI) is transforming allergy and immunology (A/I) by enabling precision care through predictive analytics, multimodal data integration, and adaptive decision support. Applications span clinical practice, research, and education, from oral food challenge risk prediction to federated learning for privacy-preserving collaboration. AI-driven tools promise improved outcomes, reduced clinician burden, and enhanced patient engagement. However, responsible adoption requires attention to bias, transparency, and trust. This article outlines current capabilities, emerging innovations, and ethical considerations for integrating AI into A/I, positioning clinicians as leaders in shaping personalized, equitable care.","[""Kagan Olga"", ""Topaz Maxim"", ""Connor MaryAnn"", ""Gardner James""]",Immunology and allergy clinics of North America,2026-09-17,10.1016/j.iac.2026.08.006,"[""Journal Article"", ""Review""]","[""Humans"", ""Precision Medicine"", ""Artificial Intelligence"", ""Allergy and Immunology"", ""Hypersensitivity"", ""Federated Learning""]", 43041782,https://pubmed.ncbi.nlm.nih.gov/43041782/,Machine learning and artificial intelligence in lung transplantation.,"As artificial intelligence (AI)-based applications become increasingly embedded in daily medical practice, it is important that healthcare institutions and providers become familiar with these tools. The integration of machine learning (ML) and AI into routine healthcare delivery means that future healthcare workers will require training to ensure appropriate application of these models. While most clinicians are familiar with traditional biostatistical approaches to medical research, the rapid development in machine learning models has likely left most practitioners underprepared to understand these new AI-based models. This review highlights fundamental concepts in ML/AI modeling, summarizes research published within lung transplantation, and explores potential areas of future research and development.","[""Babinski Piotr"", ""Warrior Krishnan"", ""Betensley Alan"", ""Joudi Anthony"", ""Misharin Alexander"", ""Scott Budinger G R"", ""Szczurek Ewa"", ""Bagci Ulas"", ""Subramani Mrinalini Venkata""]",JHLT open,2026-10-16,10.1016/j.jhlto.2026.100708,"[""Journal Article"", ""Review""]",,PMC13771752 42976586,https://pubmed.ncbi.nlm.nih.gov/42976586/,Bayesian nonparametric survival analysis via deep Dirichlet process.,"Time-to-event data analysis has attracted growing interest across various domains. A major challenge lies in the inherent incompleteness of such data, with right censoring being the most common form. Although Cox's proportional hazards model is effective for traditional datasets, it often fails in modern settings where high-dimensional covariates violate the proportional hazards assumption. Additionally, conventional parametric survival models, typically based on the exponential family, impose strong assumptions and struggle to capture long-tail behavior due to their exponential decay. To address the aforementioned challenges, we propose DDPSurv, a novel deep learning framework for survival analysis that leverages a deeply parameterized Dirichlet Process (DP) mixture model. Unlike prior deep parametric approaches that rely on strong statistical assumptions, DDPSurv offers greater flexibility in modeling survival distributions by incorporating a DP mixture, enabling improved representation of heavy-tailed behavior. We provide theoretical guarantees showing that the model can closely approximate the true survival distribution with a tight concentration rate. Empirical results on standard survival benchmarks demonstrate the effectiveness of our method, while large-scale experiments on MIMIC-III and MIMIC-IV datasets further validate its scalability and clinical utility. Codes are available at https://github.com/HKU-MedAI/DDPSurv-net.","[""Xu Jiacheng"", ""Chan Tsai Hor"", ""Wang Fuying"", ""Yin Guosheng"", ""Yu Lequan""]",Neural networks : the official journal of the International Neural Network Society,2026-10-23,10.1016/j.neunet.2026.109792,"[""Journal Article""]","[""Bayes Theorem"", ""Deep Learning"", ""Survival Analysis"", ""Humans"", ""Algorithms""]", 42976585,https://pubmed.ncbi.nlm.nih.gov/42976585/,Distributed variational bayesian meta-learning over multi-agent systems.,"Learning new tasks from limited data inherently introduces uncertainty in meta-learning. While Bayesian meta-learning aims to enhance traditional algorithms by quantifying uncertainty in model parameters, existing methods are centralized and cannot handle decentralized data. In this paper, we present the Distributed Variational Bayesian Meta-Learning (DVBML) framework, the first to enable distributed uncertainty quantification for decentralized meta-learning. Unlike deterministic approaches like Dif-MAML, DVBML models meta-parameters as probability distributions. We propose a novel gradient-diffusion hybrid algorithm that combines local stochastic gradient variational Bayes steps with a global diffusion-based consensus strategy, allowing agents to collaboratively learn and propagate both the mean and variance of task parameters. Rigorous theoretical analysis proves that this strategy guarantees convergence and achieves a linear speedup O(1/NT) in variance reduction compared to non-cooperative baselines. Extensive evaluations on few-shot regression and classification benchmarks show that DVBML matches the accuracy of centralized Bayesian meta-learning and delivers reliable, well-calibrated uncertainty estimates. It significantly outperforms non-cooperative baselines and quantitatively surpasses deterministic distributed methods in calibration metrics, maintaining remarkable robustness even under imbalanced data settings.","[""Zhang Xianyang"", ""Kang Weijie"", ""Li Ying"", ""He Bing"", ""Huang Honglan""]",Neural networks : the official journal of the International Neural Network Society,2026-10-27,10.1016/j.neunet.2026.109805,"[""Journal Article""]","[""Bayes Theorem"", ""Algorithms"", ""Uncertainty"", ""Machine Learning"", ""Humans""]", 42976584,https://pubmed.ncbi.nlm.nih.gov/42976584/,Structure-constrained language-assisted diffusion model for unpaired low-dose computed tomography angiography reconstruction.,"The application of iodinated contrast media (ICM) improves the sensitivity and specificity of computed tomography (CT) for a wide range of clinical indications. However, high ICM doses can cause problems such as kidney damage and life-threatening allergic reactions. Deep learning methods can generate CT images of normal-dose ICM from low-dose ICM, reducing the required dose while maintaining diagnostic power. However, existing methods struggle to achieve accurate enhancement with incompletely paired images, mainly because of the limited ability of the model to recognize specific structures. To overcome this limitation, we propose a Structure-constrained Language-assisted Diffusion Model (SLDM), a unified medical generation model that integrates structural synergy and spatial intelligence. First, the structural prior information of the image is effectively extracted to constrain the model inference process, thus ensuring structural consistency in the enhancement process. Subsequently, a language-assisted strategy with spatial intelligence is introduced, which integrates the functions of visual perception and spatial reasoning, thus prompting the model to achieve accurate enhancement. Finally, the subtraction angiography enhancement module is applied, which serves to improve the contrast of the ICM region to suitable interval for observation. Qualitative analysis of visual comparison and quantitative results of several metrics demonstrate the effectiveness of our method in angiographic reconstruction for low-dose contrast medium CT angiography.","[""Zhang Genyuan"", ""Wang Zihao"", ""Gao Zhifan"", ""Xu Lei"", ""Zhou Zhen"", ""Yu Haijun"", ""Zhang Jianjia"", ""Liu Xiujian"", ""Zhang Weiwei"", ""Wang Shaoyu"", ""Fu Huazhu"", ""Liu Fenglin"", ""Wu Weiwen""]",Neural networks : the official journal of the International Neural Network Society,2026-10-29,10.1016/j.neunet.2026.109801,"[""Journal Article""]","[""Humans"", ""Computed Tomography Angiography"", ""Contrast Media"", ""Deep Learning"", ""Image Processing, Computer-Assisted""]", 42971063,https://pubmed.ncbi.nlm.nih.gov/42971063/,FedRSFL: Federated recommender systems with feature-level user data sharing.,"Federated recommender systems have made great progress in recent years, and they can protect the privacy of user data while maintaining recommendation performance. However, most existing federated recommender systems require every user to keep all data strictly on the local device, and they overlook the users who care less about privacy and are willing to share data with the server for better recommendation. To address this, a few recent studies explore federated recommender systems with user data sharing, which allow users to share their data with the server. However, these methods mainly focus on interaction-level user-item data, which is often sparse, and they overlook the finer-grained feature-level data. To address this problem, this paper explores a federated recommender system framework with feature-level user data sharing, termed FedRSFL, which allows users to share part of their features with the server. However, training on the shared data is non-trivial: the same shared interaction produces two inconsistent training signals, because a client learns it from the complete local data with its private user embedding while the server learns it from the shared part alone with a separate user embedding, and directly aggregating the two models mixes unaligned parameters and hurts the global model. FedRSFL therefore couples the two sides through a bidirectional preference distillation, in which the clients teach the server the complete preference that the shared data misses, while the server teaches the clients the cross-user pattern that a single client cannot see, so that the shared data improves recommendation and every user embedding stays on its device. Experiments on three public datasets demonstrate that feature-level user data sharing improves the recommendation accuracy, with an average gain of about 3.8% in HR@10 over the strongest baseline on each dataset.","[""Yang Chaoqun"", ""Nguyen Quoc Viet Hung"", ""Yin Hongzhi"", ""Qu Liang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-29,10.1016/j.neunet.2026.109810,"[""Journal Article""]","[""Information Dissemination"", ""Federated Learning"", ""Humans"", ""Algorithms"", ""Computer Security"", ""Privacy"", ""User-Computer Interface""]", 42966017,https://pubmed.ncbi.nlm.nih.gov/42966017/,Breathing life into activation functions: the SWIn framework for sustained nonlinearity.,"Activation functions serve as the primary mechanism for introducing nonlinearity into deep learning models, playing an indispensable role in both model training and inference. By incorporating nonlinear transformations, they significantly enhance the representational capacity of neural networks, thereby improving overall model performance, particularly for complex tasks - such as Computer Vision (CV) and Natural Language Processing (NLP). However, most mainstream activation functions suffer from a common limitation: they tend to degenerate into linear-like behaviors when inputs assume extremely large or small values, consequently impairing their ability to model complex, nonlinear relationships across all non-saturated input domains. To address this limitation, this paper proposes a novel Sinusoidal Wave Integration (SWIn) framework. The core innovation lies in superimposing adjusted sinusoidal wave onto the activation function, which ensures that the composite function preserves the original monotonicity and global trend of the base activation function while guaranteeing persistent nonlinear characteristics within all non-plateau regions of the continuous base activation functions. We conducted extensive evaluations comparing standard activation functions with their SWIn-enhanced counterparts on image classification (CIFAR-100 and ImageNet-1K), machine translation (WMT14 DE-EN, CS-EN, RU-EN) tasks. Experimental results demonstrate that the SWIn framework yields significant performance improvements for continuous activation functions such as SiLU, Mish, and GELU. This is attributed to the enriched function representation and improved gradient propagation introduced by the controlled oscillations. In contrast, for inherently non-smooth functions like PReLU and SELU, the SWIn transformation tends to amplify existing gradient discontinuities, resulting in marginal or no performance gains.","[""Ran Yu"", ""Zhang Li"", ""Yuan Shi""]",Neural networks : the official journal of the International Neural Network Society,2026-10-28,10.1016/j.neunet.2026.109784,"[""Journal Article""]","[""Nonlinear Dynamics"", ""Respiration"", ""Humans"", ""Deep Learning"", ""Neural Networks, Computer""]", 42966015,https://pubmed.ncbi.nlm.nih.gov/42966015/,Enhanced P300-based brain-computer interface decoding via region-weighted multiscale temporal-spatial representation learning method.,"P300-based brain-computer interface (BCI) systems provide a vital communication channel for individuals with severe physical disabilities. While deep learning has achieved significant progress in signal processing, P300 recognition remains hindered by low signal-to-noise ratios and non-stationary dynamics. Existing deep learning approaches exhibit two primary limitations: they rely on fixed-scale temporal filters that fail to capture multi-resolution event-related potential (ERP) dynamics, and they treat electrode channels homogeneously, overlooking the functional modularity of cortical regions and the interaction between local and global network patterns. To address the above issue, we propose the Region-Weighted Multiscale Temporal-Spatial Neural Network (RWMTS-Net). The model integrates three key components: a region partition module that groups physiologically aligned electrodes, a multiscale temporal-spatial module that captures multi-resolution ERP dynamics across functional brain regions, and an adaptive channel attention mechanism that dynamically couples local dynamics with global network features. Evaluated on a public dataset and a self-collected dataset under participant-dependent conditions, RWMTS-Net significantly surpassed state-of-the-art baselines (including EEGNet, ShallowNet, EEGInception, PPNN, and DeepNet). It achieved superior average character recognition accuracies of 98.93% and 91.79% on the respective datasets, along with enhanced information transfer rates (ITRs), higher AUC, and lower performance variance. These results demonstrate that RWMTS-Net effectively resolves key spatio-temporal representation bottlenecks, offering a highly accurate and robust solution for BCI decoding.","[""Li Shurui"", ""Daly Ian"", ""Qin Ke"", ""Wang Xingyu"", ""Cichocki Andrzej"", ""Jin Jing""]",Neural networks : the official journal of the International Neural Network Society,2026-10-14,10.1016/j.neunet.2026.109768,"[""Journal Article""]","[""Brain-Computer Interfaces"", ""Humans"", ""Event-Related Potentials, P300"", ""Electroencephalography"", ""Representation Machine Learning"", ""Deep Learning"", ""Brain"", ""Signal Processing, Computer-Assisted"", ""Neural Networks, Computer""]", 42962425,https://pubmed.ncbi.nlm.nih.gov/42962425/,Transductive learning via validation risk minimization for semi-supervised learning.,"Transductive learning aims to recover the labels of unlabeled samples in a dataset with several labeled samples. The empirical risk minimization principle is commonly adopted in transductive learning. In this paper, the validation risk minimization principle is proposed and applied in transductive learning. The key idea is to introduce the validation strategies of supervised learning into transductive learning, leading to two learning frameworks with a general learning machine. Generally, these frameworks result in bilevel programming problems. Further, by using the least squares support vector machine as the learning machine in the two frameworks, we propose two transductive learning models which degenerate the bilevel programming problems into linear or quadratic integer programming problems. New algorithms are respectively designed to solve these problems efficiently. Preliminary experiments on synthetic, benchmark, and real-world datasets show that the proposed models achieve competitive transductive performance.","[""Lv Yan"", ""Wang Zhen"", ""Bai Lan""]",Neural networks : the official journal of the International Neural Network Society,2026-10-26,10.1016/j.neunet.2026.109803,"[""Journal Article""]","[""Supervised Machine Learning"", ""Algorithms"", ""Support Vector Machine"", ""Soft Computing"", ""Humans"", ""Least-Squares Analysis""]", 42962424,https://pubmed.ncbi.nlm.nih.gov/42962424/,Moving beyond graph homophily: Disentangled node representation for graph neural networks.,"Graph Neural Networks (GNNs) have demonstrated strong performance on graph-structured data through neighborhood aggregation. However, they often underperform on heterophilic graphs and suffer from over-smoothing, leading to suboptimal representations. We attribute these limitations to the lack of view-specific factorization in current node representations, which fail to reflect the multifaceted nature of node features. In this paper, we propose a novel node representation paradigm that decomposes node features into three complementary perspectives: label context, semantic, and topology. We observe that the relative importance of these perspectives varies not only across datasets but also among nodes within a single graph. To address this, we introduce RDGCN (Representation-Disentangled Graph Convolutional Network), which employs view-specific encoders and an adaptive fusion mechanism to learn and integrate multi-view representations dynamically. RDGCN mitigates harmful information mixing through view-specific factorization and enhances the discriminability of each encoder via targeted regularization. Notably, our framework is based on simple MLP and GNN modules yet achieves strong performance, demonstrating the power of multi-view representation learning. Moreover, our work establishes a generalizable node representation paradigm and offers a unified framework that moves beyond the binary homophily/heterophily divide. Extensive experiments on nine benchmark datasets validate the effectiveness of RDGCN, showcasing robust performance across both homophilic and heterophilic graphs.","[""Fang Shuyang"", ""Yang Zelong"", ""Cai Yintao"", ""Huang Yuqin"", ""Min Xiaoping""]",Neural networks : the official journal of the International Neural Network Society,2026-10-26,10.1016/j.neunet.2026.109799,"[""Journal Article""]","[""Graph Neural Networks"", ""Representation Machine Learning"", ""Algorithms"", ""Humans"", ""Neural Networks, Computer""]", 42951803,https://pubmed.ncbi.nlm.nih.gov/42951803/,Artificial Intelligence in Cleft and Craniofacial Care.,"Artificial intelligence (AI) is increasingly integrated into cleft and craniofacial care to facilitate objective assessment, optimize treatment planning, and improve outcome evaluation. By leveraging 3-dimensional imaging, machine learning, and automated workflows, AI addresses key clinical challenges such as anatomic variability, limited robust normative references, and reliance on subjective expertise. Recent developments enable streamlined data acquisition and automated design of surgical devices. This article reviews current applications and highlights clinically relevant advances toward integrated digital workflows and emerging digital twin concepts, with the goal of improving precision, efficiency, and accessibility across cleft and craniofacial treatment pathways.","[""Mueller Andreas A"", ""Solenthaler Barbara"", ""Schnabel Till N"", ""Lill Yoriko"", ""Nalabothu Prasad"", ""Benitez Benito K""]",Oral and maxillofacial surgery clinics of North America,2026-10-28,10.1016/j.coms.2026.09.007,"[""Journal Article"", ""Review""]","[""Humans"", ""Artificial Intelligence"", ""Imaging, Three-Dimensional"", ""Cleft Palate"", ""Cleft Lip"", ""Craniofacial Abnormalities"", ""Patient Care Planning"", ""Surgery, Computer-Assisted"", ""Machine Learning""]", 42950761,https://pubmed.ncbi.nlm.nih.gov/42950761/,Large language models for gait classification and treatment recommendation in cerebral palsy.,"BACKGROUND: Large language model (LLM)-based artificial intelligence systems are increasingly being explored for the interpretation of clinical and biomechanical data. However, their ability to classify cerebral palsy gait patterns and support treatment-related decision-making remains unclear. RESEARCH QUESTION: How accurately can commonly used LLM-based AI systems classify sagittal gait patterns in children with spastic diplegic cerebral palsy using kinematic graphs and three-dimensional skeletal gait videos, and how well do their treatment recommendations agree with expert decisions? METHODS: Fifteen children with spastic diplegic cerebral palsy, including three cases from each Rodda sagittal gait pattern, were retrospectively evaluated. ChatGPT, Gemini, and Claude independently classified each case using bilateral sagittal-plane kinematic graphs and three-dimensional skeletal gait model videos and provided treatment recommendations. Expert consensus served as the reference standard. Agreement was assessed using Cohen's and Fleiss' kappa, together with classification accuracy. RESULTS: Classification accuracy was higher in the graph-based condition than in the video-based condition. Claude showed the highest graph-based performance (73.3% accuracy; κ=0.67), while ChatGPT and Gemini each achieved 46.7% accuracy (κ=0.33). Video-based accuracy decreased to 26.7% for ChatGPT and Gemini (κ=0.08) and 20.0% for Claude (κ=0.00). Inter-model agreement was fair for graph-based classification (κ=0.31) but poor for video-based classification (κ=-0.07). Treatment recommendation agreement was generally low across models and modalities. SIGNIFICANCE: Under the tested conditions, LLMs achieved higher accuracy with kinematic graphs, although the difference cannot be attributed to input modality alone. Graph-based classification showed promising performance; however, limited treatment concordance indicates that current LLMs should be considered supportive rather than independent clinical decision-making tools.","[""Çelik Meltem"", ""Ertan-Birsel Sema"", ""Demirci Elif"", ""Doğan Osman"", ""Inan Muharrem""]",Gait & posture,2026-10-26,10.1016/j.gaitpost.2026.110677,"[""Journal Article""]","[""Humans"", ""Cerebral Palsy"", ""Large Language Models"", ""Child"", ""Retrospective Studies"", ""Female"", ""Biomechanical Phenomena"", ""Male"", ""Gait Disorders, Neurologic"", ""Gait"", ""Generative Artificial Intelligence"", ""Gait Analysis""]", 42950612,https://pubmed.ncbi.nlm.nih.gov/42950612/,A Lagrangian-inspired force-based loss for deep metric learning.,"Deep metric learning objectives typically define desired embedding distributions while leaving interaction directions and magnitudes implicit. We introduce a Lagrangian-inspired force-first framework prescribing and integrating conservative embedding-space interactions into a scalar objective. Under a fixed interaction configuration, the objective's negative gradient recovers the prescribed field. The framework identifies and quantifies three force-design principles: intrinsic hard mining, margin control through anchor-adaptive interaction ranges, and cardinality-weighted force balancing. Their force-based realization defines the potential energy metric loss, while an annulus-scaled projection preserves radial variation and bounds embedding norms. In frozen-encoder linear evaluation, our loss achieved the highest observed top-1 accuracies among compared objectives: 96.04%, 77.61%, and 85.94% on CIFAR-10, CIFAR-100, and ImageNet-100, respectively, with gains over the supervised contrastive loss reaching 2.88 percentage points. On long-tailed CIFAR-100 (imbalance ratio 200), accuracy was 55.54% versus 33.43% for the supervised contrastive loss, a 22.11-percentage-point gain. At this imbalance, the scale-independent ratio of within-class dispersion to between-centroid distance was lower overall (0.7387 versus 0.8372) and among the ten rarest classes (1.0607 versus 1.3000), indicating better-preserved class-level organization under cardinality skew. In class-disjoint fine-grained retrieval, our loss achieved the highest observed Recall@K at every evaluated depth on CUB-200-2011 and Cars196, with a minimum 4.07-percentage-point margin over the corresponding second-best result, demonstrating effective transfer to held-out categories. The annulus-scaled projection improved all five evaluated Euclidean metric-learning objectives over ℓ2 normalization by 1.91-7.16 percentage points. These results establish force-first objective construction as a principled and empirically effective route from desired embedding interactions to metric-learning objectives.","[""Cheon Byung Do"", ""Yu Jeongmin"", ""Kim Jungeun"", ""Kim Ha Young""]",Neural networks : the official journal of the International Neural Network Society,2026-10-23,10.1016/j.neunet.2026.109798,"[""Journal Article""]","[""Deep Learning"", ""Algorithms"", ""Humans""]", 42945866,https://pubmed.ncbi.nlm.nih.gov/42945866/,CRP-MIL: A channel re-partitioning based multiple instance learning method for whole slide image classification.,"Whole Slide Images (WSIs) possess ultra-high resolution and contain abundant pathological information, leading to highly complex modeling processes. Current Multiple Instance Learning (MIL) models for WSI analysis suffer from two major limitations: (1) inadequate modeling of contextual relationships between tumor regions and neighboring image patches, and (2) intrinsic constraints of the MIL framework itself. WSIs must be partitioned into multiple independent image patches, making the effective modeling of global semantic information a substantial challenge in this field. To address these issues, this study proposes a WSI analysis method based on Channel Re-Partitioning (CRP) to mitigate the limitations of MIL in local-global information modeling. This method captures local relational information along the channel dimension and establishes global relational information across different channels. Unlike previous methods focusing mainly on patch-level relationships, CRP adopts a channel-centric approach to capture intra- and inter-channel dependencies, thereby enabling more effective integration of local and global information. Extensive experiments on three public datasets demonstrate that CRP-MIL achieves superior performance compared with state-of-the-art MIL methods. Moreover, CRP-MIL consistently maintains its advantage across different feature extraction backbones.","[""Xie Yining"", ""Yin Wenchao"", ""Liu Zequn"", ""Qi Haochen"", ""Zhao Jing"", ""Hu Xiaoqing"", ""Zhang Ziyu""]",Neural networks : the official journal of the International Neural Network Society,2026-10-18,10.1016/j.neunet.2026.109786,"[""Journal Article""]","[""Multiple-Instance Learning Algorithms"", ""Humans"", ""Image Processing, Computer-Assisted""]", 42945865,https://pubmed.ncbi.nlm.nih.gov/42945865/,SynMatch: Rethinking consistency in medical image segmentation with sparse annotations.,"Label scarcity remains a major challenge in deep learning-based medical image segmentation. Recent studies use strong-weak pseudo supervision to leverage unlabeled data. However, performance is often hindered by inconsistencies between unlabeled images and their corresponding pseudo labels. In this work, we propose SynMatch, a novel framework that sidesteps the need for improving pseudo labels by synthesizing images to match them instead. Specifically, SynMatch synthesizes images using texture and shape features extracted from the same segmentation model that generates the corresponding pseudo labels for unlabeled images. This design enables the generation of highly consistent synthesized-image and pseudo-label pairs without introducing additional trainable components for image synthesis. We extensively evaluate SynMatch across diverse medical image segmentation tasks under semi-supervised learning (SSL), weakly-supervised learning (WSL), and barely-supervised learning (BSL) settings with increasingly limited annotations. The results demonstrate that SynMatch achieves superior performance, especially in the most challenging BSL setting. For example, it outperforms the recent strong-weak pseudo supervision-based method by 29.71% and 10.05% on the polyp segmentation task with 5% and 10% scribble annotations, respectively. The code is available at https://github.com/Senyh/SynMatch.","[""Shen Zhiqiang"", ""Cao Peng"", ""Liu Xiaoli"", ""Yang Jinzhu"", ""Zaiane Osmar R""]",Neural networks : the official journal of the International Neural Network Society,2026-10-24,10.1016/j.neunet.2026.109796,"[""Journal Article""]","[""Supervised Machine Learning"", ""Humans"", ""Deep Learning"", ""Image Processing, Computer-Assisted"", ""Algorithms""]", 42933325,https://pubmed.ncbi.nlm.nih.gov/42933325/,Convex-relaxed counterfactual optimization for non-differentiable classifiers via neural surrogates.,"Counterfactual explanations (CE) facilitate actionable recourse by identifying minimal perturbations that transition an instance across a model's decision boundary. However, generating CEs for non-differentiable and non-smooth classifiers, including common nonparametric models such as kNN, random forests, and kernel SVMs, remains challenging because gradient information is unavailable or unreliable near irregular decision boundaries. Current model-agnostic approaches often neglect inherent structural priors, while surrogate-based methods frequently suffer from ""metric distortion,"" where latent feature-space optimizations map to topologically invalid instances in the original input space. To resolve these theoretical impasses, we propose ECOer (Efficient Counterfactual Optimization with metric-preserving reconstruction regularization), a mathematically principled, end-to-end framework for post-hoc counterfactual generation for non-differentiable classifiers. ECOer constructs a ReLU-ReLU Single-hidden-layer Neural Network (R2SNN) as a differentiable surrogate, which we prove to possess tight universal approximation bounds for measurable decision regions. To preserve metric fidelity during latent-space search, we derive a novel metric-preserving reconstruction regularization mechanism grounded in Extreme Learning Machine (ELM) theory, ensuring that optimized features remain tied to the valid data manifold. Furthermore, we introduce a convex relaxation technique that yields a region-wise convex objective with zero relaxation gap within fixed decision regions. Extensive theoretical analysis and empirical evaluations on real-world benchmarks demonstrate that ECOer achieves a favorable trade-off between validity, proximity, plausibility, and online generation latency among the evaluated baselines. This work provides a rigorous algorithmic foundation for robust and reliable XAI in decision-critical tabular domains.","[""Xu Ao"", ""Li Zihao"", ""Wu Tieru""]",Neural networks : the official journal of the International Neural Network Society,2026-10-20,10.1016/j.neunet.2026.109793,"[""Journal Article""]","[""Neural Networks, Computer"", ""Classification Algorithms"", ""Algorithms"", ""Humans"", ""Soft Computing""]", 42933324,https://pubmed.ncbi.nlm.nih.gov/42933324/,Long-Horizon decision making via latent planning in decision transformers.,"Decision Transformers (DTs) address decision making problems through sequence modeling and have achieved surprisingly strong results. However, DTs still struggle in long-horizon tasks due to their poor planning ability. Existing works have demonstrated that subgoal prediction helps to guide DTs' decision making in complex and long-horizon tasks. However, explicit planning via subgoal prediction suffers from suboptimality, inefficiency and instability. In this paper, we present DT-Pro, a variant of DT that enhances its planning ability by integrating a natural implicit planning step into sequence modeling. Compared with explicit planning via subgoal prediction, the implicit planning works by inferring a latent plan from a structured plan space. Through this way, DT-Pro enables high-quality adaptive plan generation and efficient stepwise replanning with only a marginal increase in the computational cost. Extensive experimental results show that DT-Pro achieves strong performance on a variety of widely used control and navigation benchmarks.","[""Wang Zengyang"", ""Gan Qirui"", ""Zeng Yiran"", ""Zhang Xinzhi"", ""Jiang Jiuchuan"", ""Cai Yi"", ""Chen Junying"", ""Zhao Mengchen""]",Neural networks : the official journal of the International Neural Network Society,2026-10-17,10.1016/j.neunet.2026.109763,"[""Journal Article""]","[""Decision Making"", ""Humans"", ""Reinforcement Machine Learning""]", 42925673,https://pubmed.ncbi.nlm.nih.gov/42925673/,Smartphone placement recognition during walking: External validation and performance determinants.,"The opportunity to collect movement data from smartphones for prolonged periods has opened new perspectives in the field of clinical movement analysis. However, when monitoring people's mobility in free-living conditions, smartphone placement can influence the validity of the extracted digital mobility outcome. This study aimed to develop and validate an automatic smartphone placement recognition classifier and to investigate potential critical factors that can influence performance. The classifier was trained on data from 15 healthy participants using inertial signals collected from smartphones placed at six body placements during free-living walking and externally validated on over 3000 individuals from external datasets, including blind participants and patients with cardiovascular or Parkinson's disease. A decision-tree ensemble model was developed using feature subsets of increasing dimensionality, with the optimal subset comprising 50 features. Classification accuracy increased consistently when front and back pocket placements were aggregated (81.1%) and further improved when coat pocket was also included in the pocket class (88.5%), underscoring the challenge of distinguishing between fine-grained pocket placements. The best-recognised placements across the external datasets were lower back (precision: 100%, recall: 72.5%), hand (precision: 94.2%, recall: 94.5%), and the aggregated pocket class (precision: 86.7%, recall: 90.2%). Recognition accuracy changed across cohorts (0.73 - 0.85), activities (0.63 - 0.94) and speed (0.79 - 0.87), however it stayed consistent across various technological and environmental factors. Overall, this study supports the feasibility of smartphone placement recognition in walking and underscores the importance of accounting for key influencing factors when designing frameworks intended for deployment in heterogeneous real-world or clinical contexts.","[""Tasca Paolo"", ""Trentadue Giorgio"", ""Buckley Ellen"", ""Sun Shaoxiong"", ""Long Michael"", ""Ireson Neil"", ""Ciravegna Fabio"", ""Lanfranchi Vitaveska"", ""Cereatti Andrea""]",Gait & posture,2026-09-28,10.1016/j.gaitpost.2026.110649,"[""Journal Article"", ""Validation Study""]","[""Humans"", ""Smartphone"", ""Walking"", ""Female"", ""Male"", ""Adult"", ""Middle Aged"", ""Monitoring, Ambulatory"", ""Accelerometry"", ""Reproducibility of Results"", ""Aged""]", 42925623,https://pubmed.ncbi.nlm.nih.gov/42925623/,CDCLAP: Efficient distillation framework for zero-shot audio event classification.,"This paper presents a novel Cross-Modal Distillation Contrastive Language-Audio Pretraining (CDCLAP) framework to address the challenges of zero-shot audio event classification (ZAEC) under limited hardware resources. The proposed method transfers visual knowledge from an ImageNet-pretrained MobileNetV3 to a lightweight student audio network, and employs a CLAP model-comprising HTSAT and natural language processing model GPT-2-as the teacher to enhance the representational capacity and semantic generalization of the audio encoder. Furthermore, a description generation strategy using GPT-2 is introduced to augment the audio-text pairs, thereby improving model performance. Extensive evaluations on three benchmark datasets-ESC50, US8K, and TUT17 demonstrate that CDCLAP achieves superior performance compared to existing state-of-the-art methods, while maintaining significantly lower model complexity. These results validate the effectiveness of the proposed cross-modal distillation strategy in improving zero-shot recognition accuracy and robustness, offering new insights into efficient multimodal audio understanding.","[""Chen Fei"", ""Xue Hui"", ""Fang Pengfei""]",Neural networks : the official journal of the International Neural Network Society,2026-10-16,10.1016/j.neunet.2026.109764,"[""Journal Article""]","[""Humans"", ""Natural Language Processing"", ""Convolutional Neural Networks"", ""Large Language Models""]", 42925622,https://pubmed.ncbi.nlm.nih.gov/42925622/,"Tempered fractional rabies dynamics with deep neural surrogate modeling: Analysis, computation, and prediction.","Rabies is a fatal zoonotic disease for which domestic dogs constitute the main reservoir; its long and variable incubation period motivates transmission models that account for memory effects. This paper proposes a tempered fractional rabies model in which the classical time derivative is replaced by the left tempered Caputo derivative of order ξ ∈ (0, 1] and tempering parameter λ ≥ 0, so that the dynamics remain nonlocal while the influence of past states decays exponentially. We first establish the qualitative properties of the model: solutions remain nonnegative for nonnegative initial data, total populations are bounded on finite time intervals, and a unique continuous solution exists, which we prove by combining an exponential transformation with a contraction argument. We further derive the basic reproduction number R0 by the next-generation-matrix approach and prove local asymptotic stability of the disease-free equilibrium whenever R0<1, uniformly in the fractional order and the tempering parameter. A predictor-corrector product-integration scheme is then developed for the tempered system, and a mesh-refinement study confirms an empirical convergence order close to one. To accelerate repeated simulations, we train a deep neural surrogate that maps a short history window of log-transformed states to log-state increments, using a two-stage procedure consisting of one-step pretraining followed by differentiable rollout fine-tuning. Ablation experiments show that the logarithmic state representation and the increment-based targets are the decisive design choices, while rollout fine-tuning further improves worst-case robustness of free-running predictions. The selected surrogate attains high one-step accuracy (RMSE 0.4121, MAE 0.1303, R2=0.999998) and, in fully free-running closed-loop mode, reproduces unseen tempered test scenarios with mean R2=0.9990 (mean RMSE 3.32) on the benchmark cases and R2=0.9973±0.0031 over a random evaluation suite of 18 unseen tempered scenarios; accuracy degrades only in the near-untempered long-memory limit λ → 0, which we characterize explicitly. Parameter-matched LSTM and GRU baselines trained under the identical protocol do not outperform the proposed feed-forward surrogate, and in batched operation the surrogate is about two orders of magnitude faster than the numerical solver. The resulting framework provides a fast, accurate, and stable proxy of the tempered fractional rabies model that supports rapid parameter sweeps and forecasting-oriented scenario analysis.","[""Sallami Mohamed Nidhal"", ""Fredj Ouissem Ben"", ""Agarwal Praveen"", ""Naifar Omar""]",Neural networks : the official journal of the International Neural Network Society,2026-10-19,10.1016/j.neunet.2026.109788,"[""Journal Article""]","[""Rabies"", ""Animals"", ""Dogs"", ""Humans"", ""Deep Learning"", ""Computer Simulation"", ""Neural Networks, Computer"", ""Basic Reproduction Number""]", 42925621,https://pubmed.ncbi.nlm.nih.gov/42925621/,Learning long-tailed representations via diversity-enhanced contrastive learning.,"A key challenge of long-tailed classification lies in the insufficient diversity of minority class samples and their learned representations. Existing long-tailed learning methods attempt to mitigate these issues by re-sampling or re-weighting strategies, but fail to enrich the diversity of minority classes. In this manuscript, we propose diversity-enhanced contrastive learning (DECL), a novel framework designed to improve representation learning and model performance on long-tailed classification tasks. Specifically, DECL consists of three key modules: a diversity-enhancing image augmentation (DIA) module, a feature extraction module and a diversity-aware contrastive learning (DCL) module. The DIA module focuses on increasing both the diversity and quantity of minority class samples, enriching the sample space. Then, the feature extraction module extracts diverse features from the augmented samples. Finally, the DCL module ensures clearer inter-class boundaries by aligning features within the same class and maximizing separation between different classes. By integrating these modules, the proposed method effectively mitigates the challenges associated with imbalanced datasets, enhancing feature diversity, and improving model generalization. Experimental results on four prevailing long-tailed benchmark datasets demonstrate that our method outperforms state-of-the-art competitors. The code will be made public at https://github.com/ZPlus233/DECL.","[""Xu Yibo"", ""Zhao Wenyi"", ""Zheng Bin"", ""He Ziyang"", ""Zhang Weidong"", ""Yang Huihua""]",Neural networks : the official journal of the International Neural Network Society,2026-10-18,10.1016/j.neunet.2026.109790,"[""Journal Article""]","[""Representation Machine Learning"", ""Classification Algorithms"", ""Humans"", ""Algorithms"", ""Neural Networks, Computer""]", 42919725,https://pubmed.ncbi.nlm.nih.gov/42919725/,The Role of Artificial Intelligence in Oral Dysplasia and Oral Cancer.,"Oral epithelial dysplasia carries a variable risk of progression to oral squamous cell carcinoma, and conventional histopathologic grading cannot reliably predict the behavior of an individual lesion. Artificial intelligence offers a means of extracting additional information from data already obtained in routine care, including histologic images, and additional information regarding biomarkers, the tumor immune microenvironment, and genetic or epigenetic profiles. Machine learning and deep learning approaches are being investigated to estimate malignant transformation risk in dysplasia and to predict treatment response, recurrence, and survival in established oral cancer. These tools remain investigational and require external validation, but may support individualized surveillance and treatment decisions.","[""Zbarsky Steven J D"", ""Lee Andrew Y"", ""Viet Chi T""]",Oral and maxillofacial surgery clinics of North America,2026-10-23,10.1016/j.coms.2026.09.010,"[""Journal Article"", ""Review""]","[""Humans"", ""Mouth Neoplasms"", ""Artificial Intelligence"", ""Precancerous Conditions"", ""Carcinoma, Squamous Cell"", ""Cell Transformation, Neoplastic"", ""Deep Learning"", ""Disease Progression"", ""Machine Learning"", ""Soft Computing""]", 42914943,https://pubmed.ncbi.nlm.nih.gov/42914943/,Structured goal-guided reinforcement learning with large language models.,"Real-world decision-making tasks typically occur in complex and open environments, posing significant challenges to agents' exploration efficiency and long-horizon planning capabilities. Recent LLM-enhanced RL methods leverage large language models (LLMs) to guide reinforcement learning (RL) agents toward efficient exploration but typically incur high costs and suffer from instability due to a semantic mismatch between goal guidance and policy learning. In this paper, we introduce Structured Goal-guided Reinforcement Learning (SGRL), a novel framework that integrates a Structured Goal Planner and a Goal-Conditioned Action Pruner into a standard RL pipeline to guide exploration efficiently while preserving policy flexibility. Extensive experiments on Crafter, Craftax-Classic, and two reward-modified variants of Craftax-Classic demonstrate that SGRL achieves superior performance on challenging tasks requiring long-horizon exploration. The code is available at https://github.com/jiejieqi/SGRL.","[""Qi Yajie"", ""Wei Wei"", ""Li Lin"", ""Zhang Lijun"", ""Gao Zhidong"", ""Wang Da"", ""Song Huizhong""]",Neural networks : the official journal of the International Neural Network Society,2026-10-15,10.1016/j.neunet.2026.109769,"[""Journal Article""]","[""Reinforcement Machine Learning"", ""Large Language Models"", ""Goals"", ""Humans"", ""Decision Making"", ""Reinforcement, Psychology""]", 42914590,https://pubmed.ncbi.nlm.nih.gov/42914590/,CODER: Data-efficient activation of mathematical reasoning via difficulty-aware core-set.,"Large language models (LLMs) demonstrate remarkable performance in mathematical reasoning tasks, yet their fine-tuning process typically relies on substantial volumes of high-quality annotated data. Under conditions of limited computational resources and sparse data, they are highly susceptible to overfitting and catastrophic forgetting. Inspired by the classic textbooks used in human education from primary to secondary school, which embody a natural difficulty progression, this work suggests that constructing hierarchical data covering key knowledge nodes can effectively facilitate the activation of a model's logical reasoning capabilities with minimal sample cost. This paper proposes CODER, a data-efficient fine-tuning framework based on difficulty-aware set construction and collaborative regularisation. First, we constructed Math-Core-700 from a candidate pool spanning the full primary-to-university curriculum - textbook exercises, provincial mock examinations, publicly released competition problems and introductory university mathematics such as calculus. This dataset features a complete difficulty gradient spanning fundamental operations to advanced abstract concepts. By purposefully incorporating final-stage examination questions and competition proof problems, we facilitated the model's transition from simple pattern mimicry to enhanced logical reasoning. Secondly, we devise a collaborative fine-tuning strategy that leverages DoRA to decouple the magnitude and direction of parameter updates, while integrating NEFTune to inject noise into the feature space to promote manifold distribution smoothing. Empirical results provide supportive evidence that this approach achieves 90.6%(mean) and 82.9% (pass@16) accuracy on GSM8K and AMC 23 respectively using only 700 samples, achieving competitive or superior performance compared to standard LoRA baselines trained on the MATH dataset (7.5k samples) and TAL-SCQ5K dataset (5k samples). This finding suggests that during downstream fine-tuning, data filtering strategies based on a knowledge-grounded Difficulty-Aware Core-Set can potentially mitigate the diminishing returns associated with data scale. Consequently, they enable the effective activation of model capabilities without relying on extensive labelled data, thereby exploring new pathways for popularising high-performance large model training in resource-constrained environments.","[""Wang Lei"", ""Liu GuoAo"", ""Zhao XiaoYong""]",Neural networks : the official journal of the International Neural Network Society,2026-10-19,10.1016/j.neunet.2026.109765,"[""Journal Article""]","[""Large Language Models"", ""Humans"", ""Mathematics"", ""Logic""]", 42914062,https://pubmed.ncbi.nlm.nih.gov/42914062/,Single-cell and multi-algorithm profiling identifies SNRPG as a Claudin-1-associated regulator of malignant progression in triple-negative breast cancer.,"BACKGROUND: The epithelial-cell states and molecular drivers underlying breast cancer (BC) metastasis remain incompletely defined. This study aimed to identify and validate core molecules associated with metastatic progression. METHODS: Single-cell transcriptomics, high-dimensional weighted gene co-expression network analysis, machine learning, and public cohort analyses were integrated to nominate progression-associated candidate genes. The functional relevance of SNRPG and its association with Claudin-1 expression were evaluated using clinical specimens, in vitro functional assays, dual-luciferase reporter assays, a subcutaneous xenograft model, and a tail-vein experimental lung metastasis model. RESULTS: Single-cell analysis revealed a lymph node-enriched epithelial subpopulation with elevated copy-number variation burden and stemness-related features. Multi-algorithm screening prioritized SNRPG as a recurrent candidate feature associated with adverse tumor progression. SNRPG was upregulated in breast cancer and showed high expression in triple-negative breast cancer. Higher SNRPG expression was associated with poorer outcomes in exploratory microarray-based survival analyses. SNRPG knockdown reduced malignant phenotypes in vitro, restrained subcutaneous tumor growth, and decreased thoracic bioluminescence in a tail-vein experimental lung metastasis model. In TNBC cells, SNRPG knockdown increased Claudin-1 expression and CLDN1 promoter activity, whereas concomitant Claudin-1 knockdown counteracted several phenotypic effects of SNRPG depletion in vitro and in vivo. CONCLUSIONS: SNRPG is associated with lymph node-enriched epithelial-cell states and supports malignant phenotypes in the TNBC models examined. Increased CLDN1 expression contributes to the phenotypic effects of SNRPG depletion, although the mechanism linking SNRPG to CLDN1 remains unresolved.","[""Ye Zhihao"", ""Mou Jiancheng"", ""Su Haotian"", ""Luo Jie"", ""Cheng Yingyue"", ""Liu Ke"", ""Yang Zhuotao"", ""Qian Da"", ""Tang Hongchao"", ""Xu Mingxing"", ""Mu Yuxiao"", ""Tang Yuning"", ""Gui Wenjuan"", ""Pan Youpeng"", ""Chen Yi"", ""Meng Xuli"", ""Zheng Qinghui""]",Biochimica et biophysica acta. Molecular basis of disease,2026-10-22,10.1016/j.bbadis.2026.168507,"[""Journal Article""]","[""Humans"", ""Triple Negative Breast Neoplasms"", ""Female"", ""Animals"", ""Claudin-1"", ""Mice"", ""Gene Expression Regulation, Neoplastic"", ""Disease Progression"", ""Cell Line, Tumor"", ""Single-Cell Analysis"", ""Biomarkers, Tumor"", ""Single-Cell Gene Expression Analysis""]", 42913169,https://pubmed.ncbi.nlm.nih.gov/42913169/,Boosting RVFL networks: An outlier-resistant learning framework.,"The presence of noise and outliers in real-world data often degrades the reliability of classification and regression models. Traditional approaches, including fuzzy concepts and weighting schemes, often rely on specific function choices, which may limit their effectiveness. To address these challenges, we propose an AdaBoost-based random vector functional link network with bounded loss (AD-RVFL), which combines the adaptive reweighting capability of AdaBoost with the computational efficiency of RVFL networks. Unlike the traditional exponential loss used in AdaBoost, which can amplify the influence of outliers, the proposed AD-RVFL uses a bounded exponential loss that reduces the impact of extreme samples. For regression tasks, we further develop AD-RVFL-R, which reformulates the boosting strategy by introducing a relative-error-based thresholding mechanism while retaining analytically trained RVFL base learners. In addition, to analyze the impact of exponential loss in AdaBoost-based RVFL models, we present a classical AdaBoost RVFL model (CAD-RVFL), which employs the standard exponential loss. Experimental results and statistical analysis on several real-world datasets demonstrate that the proposed AD-RVFL and AD-RVFL-R achieve improved performance compared to baseline models. Furthermore, experiments under varying levels of outlier contamination show that AD-RVFL maintains strong robustness to outliers, highlighting its effectiveness.","[""Kumari Anuradha"", ""Tanveer M"", ""Suganthan P N""]",Neural networks : the official journal of the International Neural Network Society,2026-10-01,10.1016/j.neunet.2026.109681,"[""Journal Article""]","[""Boosting Machine Learning Algorithms"", ""Neural Networks, Computer"", ""Soft Computing"", ""Humans"", ""Classification Algorithms""]", 42913168,https://pubmed.ncbi.nlm.nih.gov/42913168/,Enhancing heterogeneous multi-agent autonomous exploration in unstructured terrains via dual-layer collaborative control.,"Multi-Agent Systems (MAS) are of great significance in real-world applications, as they can effectively manage complex tasks through decentralized collaboration. Most existing studies predominantly emphasize homogeneous MAS, i.e., systems consisting of agents with identical functional capabilities and standardized behavioral protocols. However, such systems often fail to adapt effectively to dynamic or disturbed environments during autonomous cooperative exploration. To this end, this paper investigates heterogeneous MAS and introduces a Reinforcement Learning-guided heterogeneous Multi-agent Autonomous Exploration framework (RL-MAE), designed to harness the complementary advantages of diverse agent types. Specifically, the LEP module achieves dynamic distortion pose compensation through a motion correction strategy tailored for heterogeneous platforms, providing a reliable perceptual foundation for task allocation and control. Furthermore, the HDA module establishes a three-level closed-loop architecture spanning planning, navigation, and adaptive coordination, where the upper layer determines target allocation and global paths via MD-MTSP, the middle layer performs local path tracking using TD3, and the lower layer triggers trajectory adjustments via dual-mode cost evaluation only when a cost-reduction condition is satisfied, with the three levels tightly coupled through explicit information feedback. Extensive experiments, including ablation studies, demonstrate that the proposed system achieves superior robustness compared to homogeneous systems in complex unstructured terrain.","[""Zhao Kexiang"", ""Hu Bingyu"", ""Pan Yingying"", ""Guo Chi"", ""Yang Yuxiang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-17,10.1016/j.neunet.2026.109781,"[""Journal Article""]","[""Reinforcement Machine Learning"", ""Cooperative Behavior"", ""Humans"", ""Autonomous Robots"", ""Algorithms"", ""Artificial Intelligence""]", 42913167,https://pubmed.ncbi.nlm.nih.gov/42913167/,Equivariant flow matching for molecular docking guided by ligand-protein interactions.,"Molecular docking is a key technique in drug discovery, but existing deep learning-based methods still struggle to generate physically valid binding poses consistently. Post-processing with empirical force fields can partially improve physical plausibility, yet it breaks end-to-end modeling, increases computational cost, and often weakens ligand-protein interactions and leads to pose deviations. Diffusion-based docking methods have recently shown improved accuracy and physical realism, but they typically require many sampling steps, leading to low inference efficiency. Approaches based on internal coordinates reduce sampling steps but suffer from complex pipelines and high per-step computational cost. In addition, many docking models rely heavily on extensive sampling and external scoring functions to identify reliable poses, further increasing inference time. To address these challenges, we propose an equivariant flow matching approach for molecular docking guided by ligand-protein interactions in Cartesian space (IFMDock). IFMDock guides the flow matching model to generate physically plausible docking poses with favorable repeated-sampling behavior, while substantially reducing the number of inference steps and improving docking accuracy and physical validity. Experiments on the PoseBusters and DeepDockingDare benchmarks demonstrate its effectiveness and superior performance.","[""Fan Zhiguang"", ""Yang Yuedong"", ""Li Xiang"", ""Chen Hongming"", ""Fan Lili""]",Neural networks : the official journal of the International Neural Network Society,2026-10-17,10.1016/j.neunet.2026.109770,"[""Journal Article""]","[""Molecular Docking Simulation"", ""Ligands"", ""Proteins"", ""Protein Binding"", ""Deep Learning"", ""Drug Discovery""]", 42907279,https://pubmed.ncbi.nlm.nih.gov/42907279/,Artificial Intelligence in Head and Neck Reconstruction.,"Head and neck reconstruction is a technically demanding subspecialty where successful outcomes require precise restoration of facial form alongside masticatory and speech function. Recent advances in artificial intelligence (AI) and machine learning algorithms offer potential to expedite clinical workflows, flatten the surgical learning curve for head and neck surgeons, and enhance risk stratification of surgical patients. This article outlines the role of AI in automated image segmentation for preoperative virtual surgical planning and intraoperative guidance for perforator identification, assessment of skin paddle viability, and microvascular anastomosis safety. Applications in perioperative optimization and postoperative care of high-risk surgical patients are discussed.","[""Ko Yui Yin"", ""Wang Mu"", ""Su Yu-Xiong"", ""Pu Jane Jingya""]",Oral and maxillofacial surgery clinics of North America,2026-10-21,10.1016/j.coms.2026.09.011,"[""Journal Article"", ""Review""]","[""Humans"", ""Artificial Intelligence"", ""Plastic Surgery Procedures"", ""Head and Neck Neoplasms"", ""Surgery, Computer-Assisted"", ""Machine Learning""]", 42906428,https://pubmed.ncbi.nlm.nih.gov/42906428/,Enhanced DNMF-graph integrated multilevel semantic factorization for contextual named entity recognition.,"Named Entity Recognition (NER) has undergone considerable changes in the recent past driven by the growing demand for models that are more robust, context aware and capable of generalizing across diverse textual domain. However, existing state of the art approaches primarily rely on local contextual representations, and often fails to capture corpus level semantic structures, particularly in noisy or low resource settings. To address these limitations, a unified NER system is proposed in the present work, that utilizes a Deep Non-Negative Matrix Factorization (DNMF) integrated with token level graph community discovery and a Transformer- Conditional Random Fields (CRF) model. The proposed model, named as DNMF-Graph-Transformer CRF, can utilize the strengths of various representation levels. By applying DNMF to the token-context matrix, several interpretable, low-rank latent semantic factors are obtained that not only complement neural representations but also help to improve the model's generalization when it has to work with a small amount of noisy data. Next, a token interaction graph is generated to represent the global relationships among tokens. Thus, community embeddings can be extracted that give structural priors at the corpus level. These features represent semantic neighbourhoods and functional groupings of tokens. Finally, the Transformer- CRF is adapted to learn the contextual aspects of the text. The performance of the proposed model is evaluated by conducting comprehensive experiments on CoNLL-2003, OntoNotes 5.0, and WNUT 2017. The results demonstrate that the proposed model significantly outperforms existing baseline models. It is observed that the performance gains are especially significant for noisy social media text.","[""Kakati Pallabi"", ""Dandotiya Devendra""]",Neural networks : the official journal of the International Neural Network Society,2026-10-11,10.1016/j.neunet.2026.109747,"[""Journal Article""]","[""Semantics"", ""Humans"", ""Natural Language Processing"", ""Representation Machine Learning"", ""Neural Networks, Computer"", ""Algorithms""]", 42906427,https://pubmed.ncbi.nlm.nih.gov/42906427/,Robust generalized two-dimensional principal component analysis with optimal bias learning.,"Two-dimensional principal component analysis (2DPCA) performs feature extraction directly in the matrix space and has shown high efficiency in image recognition. However, existing robust 2DPCA methods still have limited ability to handle occlusion, noise, and outliers, and it remains difficult to simultaneously achieve strong discriminative ability and data recoverability. To address this issue, this paper extends the robust generalized principal component analysis model (RGPCA-OB) to the two-dimensional matrix setting and proposes model RG2DPCA-OB. The proposed method separates the projection matrix from the recovery matrix. This gives more flexible feature learning and better reconstruction. An optimal bias term is learned adaptively instead of using the conventional sample mean. This improves robustness and representation ability. In addition, generalized norm loss and row-sparse regularization are introduced to suppress outliers and remove redundant features. A truncated reweighting strategy is further used to filter severely corrupted samples and improve noise resistance. An efficient alternating iterative algorithm is developed, and the monotonic decrease and convergence of the objective function values are theoretically established. Experiments on multiple image datasets demonstrate that, under different noise levels, RG2DPCA-OB improves classification accuracy by an average of approximately 0.44 percentage points and reduces reconstruction error by an average of approximately 1.28% compared with the best competing methods.","[""Xia Xinyu"", ""Yang Hu""]",Neural networks : the official journal of the International Neural Network Society,2026-10-13,10.1016/j.neunet.2026.109762,"[""Journal Article""]","[""Principal Component Analysis"", ""Algorithms"", ""Humans"", ""Image Processing, Computer-Assisted"", ""Machine Learning"", ""Pattern Recognition, Automated""]", 42906426,https://pubmed.ncbi.nlm.nih.gov/42906426/,Enhancing out-of-distribution detection via local-image set.,"Out-of-distribution (OOD) detection is essential for ensuring the reliability and safety of machine learning systems in open-world scenarios. Most existing methods rely on global representations of the entire image to determine whether a sample is OOD, assuming that the most informative features are uniformly distributed or sufficiently captured at a coarse spatial level. A common yet underexplored failure case arises when in-distribution (ID) images contain small foreground objects and large background regions, which often leads to high false positive rates (FPRs) across various visual recognition models. We attribute this issue to inherent spatial biases, scale sensitivity, and limitations in global feature extraction that limit clear foreground-background separation. To tackle this challenge, we propose a simple yet effective selection-then-calculation strategy, which generates multiple randomly local views of an input image and selects the one with the lowest predictive entropy for final OOD scoring. This process implicitly emphasizes salient foreground regions while suppressing irrelevant background noise, thereby improving the ability to distinguish between ID and OOD inputs. Our method is model-agnostic and can be seamlessly integrated into existing OOD detection pipelines as a plug-and-play component. Experiments demonstrate consistent improvements in OOD detection performance across various architectures.","[""Chen Qichao"", ""Wu Yanqi"", ""Li Kuan"", ""Chen Zhiyuan"", ""Maul Tomas"", ""Yin Jianping""]",Neural networks : the official journal of the International Neural Network Society,2026-10-12,10.1016/j.neunet.2026.109754,"[""Journal Article""]","[""Machine Learning"", ""Image Processing, Computer-Assisted"", ""Humans"", ""Algorithms"", ""Pattern Recognition, Automated"", ""Detection Algorithms""]", 42906425,https://pubmed.ncbi.nlm.nih.gov/42906425/,Echo-DM: Ultrasound marker removal via conditional latent diffusion and region-aware fusion.,"Clinical ultrasound images often contain artificial markers, such as measurement calipers and text, to assist diagnostic interpretation and comparison. However, these markers can introduce shortcut bias in downstream automated analysis, encouraging deep learning models to rely on marker-related cues rather than clinically meaningful anatomy. Existing marker removal methods are either mask-dependent and vulnerable to error propagation, or mask-free deterministic restorers that may over-smooth ultrasound texture and perturb unaffected background regions. To address these challenges, we present Echo-DM, a framework for ultrasound marker removal via conditional latent diffusion and region-aware fusion. Echo-DM follows a generate-select-refine formulation: a DiT-based Conditional Latent Diffusion Network (CLDN) generates an intermediate restoration, and a Region-Aware Fusion (RAF) module internally predicts a soft map to selectively fuse the restoration with intact input content before residual refinement in image space under end-to-end mask-free inference. The image-space RAF is latent-module-agnostic and is shared by Echo-DM-V and Echo-DM-R, which use VAE-based and RAE-based latent modules, respectively. Extensive experiments on Echo-PAIR, a large-scale paired clinical ultrasound dataset, demonstrate superior marker removal and strong anatomical fidelity compared with representative two-stage baselines, while providing favorable quality-efficiency trade-offs across deployment settings. Code and models will be released at Echo-DM.","[""Wang Zhiwei"", ""Huang Tao"", ""Jiang Wentao"", ""Li Muyi"", ""Liu Jianxin"", ""Chen Jian"", ""Zou Jie"", ""Luo Yong"", ""Du Bo"", ""Zhang Jing""]",Neural networks : the official journal of the International Neural Network Society,2026-10-14,10.1016/j.neunet.2026.109757,"[""Journal Article""]","[""Ultrasonography"", ""Humans"", ""Deep Learning"", ""Image Interpretation, Computer-Assisted"", ""Image Processing, Computer-Assisted""]", 42906424,https://pubmed.ncbi.nlm.nih.gov/42906424/,Efficient zero-shot post-training quantization of large language models.,"Recently, Large Language Models (LLMs) have revolutionized artificial intelligence; however, their massive parameter counts create a significant barrier to deployment on memory-constrained hardware. A critical challenge for ubiquitous AI is the memory bandwidth bottleneck, as billions of weights must be moved from memory to compute units for every generated token. While Post-Training Quantization (PTQ) offers a practical path to compression, existing state-of-the-art methods still struggle with major limitations: they either rely on computationally expensive calibration data, which risks overfitting or data unavailability, or use hardware-unfriendly mixed-precision representations to handle outliers. To address these limitations, this work develops a fast, calibration-free, and uniform-precision quantization framework. Specifically, we introduce two distinct methodologies: Hybrid Group Quantization (HGQ), which employs a Dual-Ruler strategy to statistically separate outliers from common weights, and Adaptive Grouped Vector Quantization (AGVQ), which learns local codebooks to capture sub-vector geometry. We test our methods across three diverse model families (Llama, OPT, and BLOOM), with model sizes ranging from 125M to 13B parameters. We outperform state-of-the-art calibration-based baselines in quantization runtime. HGQ reduces the quantization runtime for Llama-2-13B from over 30 minutes to under 20 seconds ( ∼ 100 × speedup) while maintaining competitive perplexity. Furthermore, AGVQ achieves highly competitive zero-shot accuracy on common sense reasoning tasks on the Llama-2-13B model, closely approximating the original FP16 performance while saving approximately 74% of the memory footprint. With fused reconstruction kernels, these savings further translate into measured speedup, HGQ reaching 1.28 × the decode throughput of the FP16 baseline on a 24 GB workstation GPU. These results demonstrate the potential of statistical, calibration-free approaches for democratizing accurate and efficient LLM deployment.","[""Haq Sami Ul"", ""El-Maleh Aiman H"", ""Alsuwaiyan Ali""]",Neural networks : the official journal of the International Neural Network Society,2026-10-13,10.1016/j.neunet.2026.109750,"[""Journal Article""]","[""Large Language Models"", ""Algorithms"", ""Humans"", ""Artificial Intelligence"", ""Neural Networks, Computer""]", 42906423,https://pubmed.ncbi.nlm.nih.gov/42906423/,Fast multi-view clustering via smooth low-rank tensor learning.,"Multi-view clustering aims to learn a consistent clustering structure by utilizing complementary information across different feature views, but balancing discriminative representation learning and computational efficiency remains challenging. Existing anchor-based tensor methods exploit high-order cross-view correlations to improve representation learning. However, suppressing localized noise perturbations while preserving dominant structural components and correcting semantic inconsistency across heterogeneous views are issues that remain insufficiently addressed. In addition, some methods still rely on spectral clustering post-processing on the sample graph, which incurs extra computational and storage overhead and restricts their scalability on large datasets. To address these limitations, we propose Fast Multi-view Clustering via Smooth Low-Rank Tensor Learning (FSLR-MVC). Specifically, we first integrates smooth nonconvex low-rank regularization into anchor tensor learning, enabling the model to adaptively preserve the dominant structural information of the anchor graphs while suppressing local noise perturbations. Next, anchor-guided geometric regularization further incorporates the geometric relationships among anchors into the representation learning process, thereby improving the structural reliability of the learned anchor graphs. Subsequently, we employ cross-view semantic alignment learning to map the view-specific anchor graphs into a shared consensus representation, thereby reconciling the heterogeneous semantics across different views. Finally, FSLR-MVC directly performs k-means on the consensus representation without constructing an n×n sample affinity matrix or conducting corresponding spectral decomposition. The overall computational complexity of the algorithm grows linearly with the number of samples, providing favorable scalability for large-scale datasets. Experimental results on several benchmark datasets show that FSLR-MVC achieves competitive clustering performance while maintaining high computational efficiency.","[""Zhang Zhao"", ""Zhang Shuguang"", ""Lu Gui-Fu""]",Neural networks : the official journal of the International Neural Network Society,2026-10-10,10.1016/j.neunet.2026.109751,"[""Journal Article""]","[""Clustering Algorithms"", ""Representation Machine Learning"", ""Semantics""]", 42906420,https://pubmed.ncbi.nlm.nih.gov/42906420/,PhyDiff: Physics-guided latent diffusion model for sparse-view CT projection interpolation.,"Sparse-view Computed Tomography (SVCT) is an important strategy for minimizing radiation exposure and accelerating industrial inspection, yet it suffers from severe streak artifacts under angular undersampling. Existing sinogram completion methods often process detector rows independently, failing to exploit cross-row correlations to maintain the structural consistency required for high-fidelity reconstruction. To address this, we propose PhyDiff, a physics-guided latent diffusion framework that reformulates sparse-view projection completion as an angle-conditioned sequence interpolation. PhyDiff incorporates the known acquisition angles to condition cross-view alignment and exploit geometric correlations between adjacent projections. Specifically, a Physical-Geometric Autoencoder (PGAE) decomposes projection synthesis into geometry-guided warping and residual refinement. The warping model captures the dominant translation and scale variation between neighboring views, while the residual branch recovers spatially varying residuals that cannot be represented by the low-order transformation. Furthermore, a Spectral-Aware Latent Diffusion (SALD) model introduces angle-conditioned spectral modulation into the generative process to preserve fine structural details. Experimental results on simulated and measured datasets demonstrate that PhyDiff consistently improves projection fidelity under different acquisition geometries.","[""Duan Shuo"", ""Wu Xing"", ""Zhu Yimin"", ""Zhang Xinyuan"", ""Wei Xing"", ""Qian Quan"", ""Song Jun"", ""Dai Jianbiao""]",Neural networks : the official journal of the International Neural Network Society,2026-10-15,10.1016/j.neunet.2026.109744,"[""Journal Article""]","[""Autoencoder"", ""Tomography, X-Ray Computed"", ""Humans"", ""Image Processing, Computer-Assisted"", ""Algorithms"", ""Artifacts""]", 42896472,https://pubmed.ncbi.nlm.nih.gov/42896472/,A single-loop bilevel deep learning method for optimal control of obstacle problems.,"Optimal control of obstacle problems arises in a wide range of applications and is computationally challenging due to the nonsmoothness, nonlinearity, and bilevel structure of the underlying problems. Classical numerical methods typically rely on mesh-based discretizations and require solving a sequence of computationally expensive subproblems, while deep learning approaches for this class of problems remain largely unexplored. In this work, we propose a bilevel deep learning framework whose core training algorithm is the computationally efficient Single-Loop Stochastic First-Order Bilevel Algorithm (S2-FOBA), which avoids nested optimization, uses only first-order information, and does not rely on restrictive lower-level uniqueness assumptions. The proposed method is mesh-free, scalable to high-dimensional settings and complex domains, and avoids the repeated solution of discretized subproblems. It employs constraint-embedding neural networks to approximate both the state and the control, thereby enforcing admissibility by construction while preserving the intrinsic bilevel structure of the problem. We further analyze the convergence behavior of S2-FOBA under mild conditions. Numerical experiments on benchmark problems, including two-dimensional examples with regular and irregular obstacles, a five-dimensional example, and a problem posed on a complex domain, demonstrate the effectiveness, efficiency, and robust constraint enforcement of the proposed approach.","[""Song Yongcun"", ""Zeng Shangzhi"", ""Zhang Jin"", ""Zhang Lvgang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-14,10.1016/j.neunet.2026.109761,"[""Journal Article""]","[""Deep Learning"", ""Algorithms"", ""Neural Networks, Computer"", ""Soft Computing"", ""Computer Simulation""]", 42896468,https://pubmed.ncbi.nlm.nih.gov/42896468/,DA-S2RL: Disease-aware shared-specific representation learning for heterogeneous image-EHR Alzheimer's disease diagnosis.,"Combining electronic health records (EHR) with medical images is important for improving the diagnosis of Alzheimer's disease (AD). However, neuroimages and EHR differ greatly in semantic content and data distribution. Neuroimages mainly reflect brain structure and metabolic abnormalities, while EHR records clinical symptoms and behavioral changes. Directly fusing these heterogeneous modalities may mix shared disease information with modality-specific diagnostic cues, resulting in redundant and ambiguous representations. To address this problem, we propose DA-S2RL, a disease-aware shared-specific multimodal representation learning framework for AD diagnosis. DA-S2RL organizes heterogeneous image and EHR features into disease-related shared semantics and modality-specific diagnostic residuals. The core of DA-S2RL is the shared-specific cross-modal coding (SCC) module. It strengthens cross-modal semantic consistency, suppresses redundant correlations, and enables mutual refinement between shared semantics and diagnostic residuals. Before SCC, image features are integrated into a compact image representation, and EHR features are encoded into the same feature space. Experiments on ADNI and NACC demonstrate that explicitly organizing heterogeneous image-EHR representations into shared disease semantics and modality-specific residuals improves multimodal diagnostic learning under different image-EHR settings.","[""Wang Wei"", ""Shi Yu"", ""Cheng Juan"", ""Liu Yu""]",Neural networks : the official journal of the International Neural Network Society,2026-10-19,10.1016/j.neunet.2026.109760,"[""Journal Article""]","[""Alzheimer Disease"", ""Humans"", ""Representation Machine Learning"", ""Electronic Health Records"", ""Neuroimaging"", ""Semantics"", ""Brain""]", 42896432,https://pubmed.ncbi.nlm.nih.gov/42896432/,PEGylated fucoxanthin liposomes remodel G0/G1 cell cycle homeostasis and suppress gallbladder cancer progression via modulation of NR2F1-mediated signaling networks.,"PURPOSE: Gallbladder cancer (GBC) is a highly aggressive and refractory biliary malignancy with disordered proliferative homeostasis, rapid malignant progression and scarce effective clinical therapies. Marine fucoxanthin possesses prominent anti-tumor activity, whereas its poor water solubility and low bioavailability limit application, and its core biomolecular mechanism against GBC remains poorly clarified. This study aimed to construct PEGylated fucoxanthin liposomes and elucidate their anti-tumor mechanism targeting biomolecular dynamics and core regulatory targets in GBC. METHODS: The anti-proliferative, anti-migratory, pro-apoptotic and cell cycle-regulatory effects of fucoxanthin-liposomes were verified using 2D monolayer and physiologically biomimetic 3D GBC spheroid models. Integrative transcriptomic analysis, multi-algorithm machine learning screening, RT-qPCR and immunofluorescence validation were performed to screen core functional targets. Molecular docking and 100 ns molecular dynamics simulations were further conducted to confirm the stable binding characteristics between fucoxanthin and the target protein. RESULTS: Fucoxanthin-liposomes exerted significant dose-dependent inhibitory effects on GBC proliferation and migration, induced mitochondrial structural damage, extensive cell apoptosis, and robust G0/G1 cell cycle arrest. Multi-omics combined with machine learning identified NR2F1 as the pivotal target, whose expression was markedly upregulated after fucoxanthin-liposomes treatment. Fucoxanthin could stably bind NR2F1 through hydrogen bonding and hydrophobic interactions with excellent thermodynamic stability. CONCLUSION: PEGylated fucoxanthin liposomes restored G0/G1 quiescence-dominated cellular homeostasis and reversed pathological proliferative imbalance in GBC via transcriptional activation of NR2F1. Upregulated NR2F1 remodeled tumor proliferative homeostasis by hierarchically modulating PI3K/AKT/mTOR and JAK2/STAT3 signaling networks, and ultimately restrained GBC clonal proliferation and invasive progression.","[""Cao Jun-Feng"", ""Hang Kuan"", ""Zhang Xiao"", ""Xia Qingjie"", ""Xia Zengliang"", ""Liao Dunshui"", ""Rong Cheng"", ""Li Kezhou""]",Biomaterials advances,2026-10-15,10.1016/j.bioadv.2026.215251,"[""Journal Article""]","[""Xanthophylls"", ""Humans"", ""Liposomes"", ""Gallbladder Neoplasms"", ""Signal Transduction"", ""Polyethylene Glycols"", ""Cell Proliferation"", ""Cell Line, Tumor"", ""Apoptosis"", ""Homeostasis"", ""Cell Movement"", ""Resting Phase, Cell Cycle"", ""Molecular Docking Simulation"", ""Disease Progression""]", 42892271,https://pubmed.ncbi.nlm.nih.gov/42892271/,Unified dual-decoder autoencoder for federated multi-modal clustering with arbitrary modality missingness.,"Federated multi-modal clustering (FedMMC) aims to collaboratively group data across multiple clients without sharing raw data. However, most existing methods still adopt restrictive client assumptions, considering clients as single-modal, fully multi-modal, or only a mixture of these two types, which is still unrealistic in practice. In real federated scenarios, clients often possess arbitrary subsets of modalities, leading to client-level arbitrary modality missingness. To address this issue, we propose a Unified Dual-Decoder Autoencoder framework for FedMMC. It uses a dual-decoder architecture to disentangle modality-shared and modality-specific information, and introduces modality-specific one-hot indicator vectors with an evaluator-guided fusion module to adaptively integrate heterogeneous modality subsets. In addition, we design a client-configuration-guided modality-missing simulation strategy with sequential parameter propagation to improve robustness under arbitrary modality-missing settings, and develop a mutual information-guided weighted aggregation strategy for global optimization. Extensive experiments on four datasets demonstrate that our method consistently outperforms existing methods in heterogeneous federated settings. The source code is available at https://github.com/xzy1095/UUDA.","[""Xu Zhenhao"", ""Chen Huazhu"", ""Wang Weiwei"", ""He Ruiqiang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-13,10.1016/j.neunet.2026.109746,"[""Journal Article""]","[""Autoencoder"", ""Federated Learning"", ""Clustering Algorithms"", ""Humans"", ""Cluster Analysis"", ""Algorithms""]", 42887167,https://pubmed.ncbi.nlm.nih.gov/42887167/,ConsistDiff: Training-free reinforcement learning for semantically consistent text-to-image generation.,"Diffusion models have become a leading paradigm for text-to-image generation owing to their promising capability to produce photo-realistic and diverse images from text prompts. Despite these advances, state-of-the-art models often fail to maintain faithful semantic consistency, particularly when handling complex descriptions with multiple entities and attributes. These shortcomings manifest as object omission, attribute leakage, or inaccurate binding between textual concepts and visual elements. To address these challenges, particularly for deploying generative models on resource-constrained devices, we propose ConsistDiff, an inference-time optimization framework that enhances Stable Diffusion without requiring computationally expensive fine-tuning. By formulating the denoising process as an online reinforcement learning task, we dynamically adjust token weights to improve semantic alignment. Specifically, we integrate a context-aware attention mechanism to refine inter-token relations by leveraging character-level spatial distances. This context-aware design provides a fine-grained structural signal that sharpens attribute-entity binding, suppresses spurious long-range couplings, and adds negligible computational overhead. Further, we formulate prompt-specific token reweighting as a reinforcement-learning problem and optimize it with an attention-separation reward that explicitly penalizes cross-entity attention overlap while preserving the saliency of key objects. Extensive experiments on multiple benchmark datasets demonstrate that ConsistDiff consistently improves semantic alignment between the input text prompts and the generated images while maintaining highly efficient inference as well as mitigating common issues such as attribute leakage and entity inconsistency.","[""Wang Junwei"", ""Zhang Kaibing"", ""Li Pengfang"", ""Liu Meng"", ""Li Xiaoyan""]",Neural networks : the official journal of the International Neural Network Society,2026-10-10,10.1016/j.neunet.2026.109734,"[""Journal Article""]","[""Reinforcement Machine Learning"", ""Semantics"", ""Humans"", ""Generative Artificial Intelligence"", ""Algorithms""]", 42887166,https://pubmed.ncbi.nlm.nih.gov/42887166/,Architecture-agnostic knowledge distillation for machine-learning interatomic potentials.,"Atomistic foundation models can achieve near-ab-initio accuracy and strong transferability across molecular and materials systems, but their large parameter counts lead to high inference latency and memory cost, limiting long-time-scale molecular dynamics and resource-constrained deployment. In many practical studies, only a narrow chemical subdomain is required, motivating compact students that preserve the teacher's physical priors. Existing distillation for Machine-Learning Interatomic Potentials (MLIPs) is still limited by architecture mismatch or incomplete geometric supervision. We propose an architecture-agnostic framework, Joint Atomic Energy-Energy Hessian Distillation, that requires no architecture-specific feature projection. The method combines atomic-energy alignment, which transfers local decomposition priors, with Hessian alignment, which transfers local curvature of the potential energy surface. These two signals are complementary and jointly regularize both decomposition consistency and geometric stability. Under matched training settings, our method consistently improves energy accuracy over a strong Hessian-distillation baseline (average gain: 48.3% on SPICE and 6.7% on MPtrj) while preserving force accuracy. Across teacher-student pairs, it also reduces parameter counts by 79%-98%, enabling fast and deployment-friendly specialized MLIPs for targeted chemical subdomains.","[""Wang Chen"", ""Hu Siyu"", ""Tan Guangming"", ""Jia Weile""]",Neural networks : the official journal of the International Neural Network Society,2026-10-09,10.1016/j.neunet.2026.109742,"[""Journal Article""]","[""Machine Learning"", ""Neural Networks, Computer""]", 42880040,https://pubmed.ncbi.nlm.nih.gov/42880040/,KAPI-LM: kernel-adaptive physics-informed learning machines for PDEs.,"Physics-informed neural networks (PINNs) have emerged as mesh-free approaches for solving partial differential equations, but their learned representations are often difficult to interpret. RBF-kernel-based formulations such as sparse, physics-based, and interpretable neural networks (SPINNs) make the geometry of the physical kernels explicit and show that kernels adapt coherently around dominant solution features, motivating the question of whether all kernel geometries need to remain independent. To investigate this question, we introduce kernel-adaptive physics-informed learning machines (KAPI-LMs), a general framework for solving PDEs with structured RBF representations in which a large physical kernel dictionary is organized through a smaller set of geometric variables while retaining all physical kernels and independent expansion coefficients. The resulting hypothesis class provides explicit control over the degree of geometric sharing and can be combined with different physics-informed learning strategies. To demonstrate this flexibility, we develop compressed SPINN (cSPINN) using joint geometry-coefficient optimization, compressed physics-informed extreme learning machine (cPIELM) using variable projection for coefficient-linear PDEs, and Meta-cSPINN using task-conditioned learning for parametric PDEs. We establish approximation guarantees and quantify the effect of geometric sharing. Numerical studies across linear, nonlinear, parametric, coupled, and multiscale PDEs demonstrate substantial geometric compression at fixed physical basis resolution and reveal problem-dependent limits to sharing. They also show that different optimization strategies can discover markedly different geometries within the same hypothesis class. For parametric PDEs, Meta-cSPINN supports one-shot prediction and suggests additional organization across solution families, while PDE residuals identify where geometric sharing should be released or local resolution added. KAPI-LM thus provides a physics-informed PDE-solving framework in which geometric complexity is explicit and controllable, allowing the solution representation to be interpreted, compressed, and adaptively refined according to the physics.","[""Dwivedi Vikas"", ""Srinivasan Balaji"", ""Sigovan Monica"", ""Sixou Bruno""]",Neural networks : the official journal of the International Neural Network Society,2026-10-12,10.1016/j.neunet.2026.109740,"[""Journal Article""]","[""Neural Networks, Computer"", ""Machine Learning"", ""Physics"", ""Soft Computing"", ""Extreme Learning Machines"", ""Radial Basis Function Networks""]", 42880035,https://pubmed.ncbi.nlm.nih.gov/42880035/,JTCSE: Joint tensor-modulus constraints and cross-attention for unsupervised contrastive learning of sentence embeddings.,"Unsupervised contrastive learning has attracted significant attention in natural language processing for learning sentence embeddings. Existing methods primarily constrain the orientation distribution of representations in high-dimensional semantic space via cosine-based objectives, yet the embedding vector also possesses a magnitude (norm) property. We observe that ignoring magnitude consistency between positive-pair representations limits the effectiveness of contrastive learning. To address this, we propose a magnitude-alignment training objective that explicitly constrains the norm discrepancy between positive-pair embeddings, strengthening contrastive alignment from both directional and magnitude perspectives. Furthermore, BERT-like encoders suffer from the attention sink phenomenon, whereby disproportionate attention mass is allocated to non-informative tokens at the expense of the [CLS] token that aggregates sequence-level semantics. To mitigate this, we introduce a cross-attention mechanism within a dual-encoder architecture that enhances the model's focus on [CLS] representations and improves the quality of CLS-based pooling. Building upon these two complementary ideas, we present JTCSE (Joint Tensor representation modulus constraint and Cross-attention unsupervised contrastive learning Sentence Embedding framework). We evaluate JTCSE on seven semantic textual similarity benchmarks with controlled multi-seed experiments and statistical significance tests; both the dual-encoder ensemble model and its single-encoder distilled variant achieve competitive or state-of-the-art performance. Extensive zero-shot evaluation across more than 140 downstream tasks further demonstrates that JTCSE consistently outperforms existing baselines. Source code is available at https://github.com/tianyuzong/JTCSE, saved checkpoints are available at https://huggingface.co/UCASzty.","[""Zong Tianyu"", ""Shi Bingkang"", ""Yi Hongzhu"", ""Wang Yuanxiang"", ""Zhang Zhenghao"", ""Xu Jungang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-08,10.1016/j.neunet.2026.109724,"[""Journal Article""]","[""Unsupervised Machine Learning"", ""Semantics"", ""Humans"", ""Natural Language Processing"", ""Attention""]", 42874412,https://pubmed.ncbi.nlm.nih.gov/42874412/,MCLE: Multimodal collaborative learning and enhancement with similarity-based modal purification for pretrained features.,"Multimodal recommendation has become a crucial framework for leveraging various data modalities to boost recommendation performance. However, existing methods still face several limitations when exploiting multimodal features. Relying solely on limited raw features prevents full exploitation of the rich semantic information inherent in various modalities, while unfiltered noise in multimodal data can degrade recommendation quality. Moreover, naive modal feature fusion strategies struggle to address the significant differences in preferences for different modalities among various users. To address these limitations, we propose a Multimodal Collaborative Learning and Enhancement framework (MCLE). Specifically, we utilize a pretrained language model to augment user profiles and item attributes, and propose an adaptive fusion mechanism that combines the enhanced user profile features with interaction-derived user representations, rather than simply relying on either source alone. A Similarity-based Modal Purification (SMP) module is then designed to filter out modal feature noise using learnable reference vectors that evaluate feature quality through similarity computation, producing purified multimodal representations. On this basis, a Multimodal Adaptive Feature Aggregator (MAFA) employs difference-aware attention and dynamic modal gating to achieve personalized multimodal feature integration. Extensive experiments on three public datasets demonstrate the effectiveness of our approach.","[""Wang Guozhen"", ""Shi Yuliang"", ""Cheng Lin"", ""Qi Kaiyuan"", ""Wu Dong"", ""Li Hui"", ""Chen Zhiyong""]",Neural networks : the official journal of the International Neural Network Society,2026-10-08,10.1016/j.neunet.2026.109732,"[""Journal Article""]","[""Humans"", ""Algorithms"", ""Cooperative Behavior"", ""Machine Learning"", ""Neural Networks, Computer"", ""Adaptive Algorithms""]", 42874411,https://pubmed.ncbi.nlm.nih.gov/42874411/,Bio-inspired zero-shot learning with hierarchical fault-trait knowledge transfer for cross-domain diagnosis.,"Fault diagnosis methods for complex mechanical systems often face challenges in adapting to unseen fault types and varying operating conditions. Existing techniques typically suffer from limited cross-domain adaptability and weak zero-shot recognition capabilities. To overcome these limitations, this paper proposes a bio-inspired zero-shot learning framework that incorporates hierarchical fault-trait knowledge transfer and domain adaptation. A biomimetic network is designed to quantify domain discrepancies and generate transferable domain factors, using a hierarchical fault-trait description framework inspired by the differentiation process from general to specialized features. This approach enables fault diagnosis of previously unseen fault types by leveraging labeled data from the source domain only. The results demonstrate the effectiveness of the proposed method for cross-domain zero-shot diagnosis and suggest its potential applicability to complex mechanical systems.","[""Yang Qiuyu"", ""Lin Yuyi"", ""Ruan Jiangjun"", ""Han Yun"", ""Xie Jingyi"", ""Xue Xue""]",Neural networks : the official journal of the International Neural Network Society,2026-10-02,10.1016/j.neunet.2026.109702,"[""Journal Article""]","[""Neural Networks, Computer"", ""Algorithms"", ""Biomimetics"", ""Machine Learning"", ""Soft Computing"", ""Humans""]", 42874410,https://pubmed.ncbi.nlm.nih.gov/42874410/,TMRKP: Trustworthy MRI-based radiology knowledge-aware prediction framework for stroke-associated pneumonia and hemorrhagic transformation.,"Prediction of stroke-associated pneumonia (SAP) and hemorrhagic transformation (HT) is critical for improving the prognosis of acute ischemic stroke patients. However, current brain magnetic resonance imaging (MRI)-based models narrowly focus on generic visual patterns, and frequently overlook the radiology-informed knowledge critical for predicting SAP and HT. To address this limitation, we propose a trustworthy MRI-based radiology knowledge-aware prediction (TMRKP) framework equipped with radiology knowledge-aware masked modeling (RKMM) and evidential multi-sequence fusion (EMF). Specifically, TMRKP first organizes four MRI sequences into a 2 × 2 super-sequence configuration, mimicking the diagnostic perspective of radiologists who simultaneously analyze multiple imaging modalities. Second, to introduce radiology knowledge into representation learning, we focus on white matter hyperintensities (WMH), a well-established biomarker of chronic small-vessel disease and long-term ischemic injury that is clinically relevant to SAP and HT. During pre-training, RKMM segments WMH masks from the MRI sequence to steer the model toward WMH regions, enabling it to learn meaningful representations of these clinically relevant pathological areas. Third, for SAP and HT prediction, EMF models the predictive uncertainty using evidence derived from both the super-sequence and the WMH-aware masked super-sequence, and fuses the resulting multi-sequence evidence based on Dempster-Shafer theory (DST). Extensive experiments on the collected Jiangdu dataset, together with independent external validation on the Zhongda dataset, demonstrate the superiority and generalizability of TMRKP over the state-of-the-art models.","[""Zhao Chenyang"", ""Yu Yaoyao"", ""Xia Tianyi"", ""Yuan Mengxuan"", ""Yang Yunjun"", ""Hu Chuanfei""]",Neural networks : the official journal of the International Neural Network Society,2026-10-08,10.1016/j.neunet.2026.109726,"[""Journal Article""]","[""Humans"", ""Magnetic Resonance Imaging"", ""Stroke"", ""Pneumonia"", ""Representation Machine Learning""]", 42874408,https://pubmed.ncbi.nlm.nih.gov/42874408/,Topology-aware dynamic co-evolution for robust graph learning.,"Graph Neural Networks (GNNs) have achieved remarkable success in learning from graph-structured data. However, recent studies have revealed their vulnerability to adversarial attacks, where imperceptible perturbations to the graph structure can easily mislead predictions. Existing defense methods predominantly rely on global heuristic strategies, which often lead to over-smoothing in deep architectures and fail to capture localized, fine-grained structural anomalies. In this paper, we propose Topology-Aware Dynamic Co-evolution (TADC), a novel framework that reformulates graph defense as a closed-loop optimization process. Unlike static purification paradigms, TADC introduces a closed-loop dynamic co-evolution framework, in which graph topology and model parameters are jointly optimized through validation-guided meta-gradients, enabling continuous and adaptive structural refinement during training. To further ensure the reliability of topology updates, we identify the topology-agnostic limitation of conventional meta-gradients and propose a topology-aware gradient calibration mechanism, which incorporates structural priors to guide gradient directions toward more plausible and semantically consistent graph modifications. Concretely, TADC first localizes vulnerable nodes via a structural-semantic consistency probe and constructs an initialized graph through topology-aware refinement. It then performs structure decoupling to disentangle reliable and potentially adversarial edges into distinct channels. Building upon this decomposition, a fully differentiable attention-based architecture DeepGCN enables end-to-end optimization of the co-evolution process. Extensive experiments on multiple benchmark datasets demonstrate that TADC consistently outperforms state-of-the-art methods across various attack settings, with particularly pronounced advantages under strong adversarial perturbations and high perturbation rates. Our code is available at https://github.com/robustgcn/TADC.","[""Zhang Tingting"", ""Lu Shuxia"", ""Wang Mengru"", ""Liu Bin""]",Neural networks : the official journal of the International Neural Network Society,2026-10-14,10.1016/j.neunet.2026.109736,"[""Journal Article""]","[""Graph Neural Networks"", ""Algorithms""]", 42869727,https://pubmed.ncbi.nlm.nih.gov/42869727/,Blepharoplasty in the Artificial Intelligence Era.,"Blepharoplasty and related periorbital procedures sit at the interface of functional reconstruction, esthetic surgery, ophthalmology, facial plastic surgery, and oral and maxillofacial surgery. Because eyelid and periorbital changes can be captured in photographs, videos, and 3-dimensional imaging, artificial intelligence has growing relevance for objective measurement, screening, planning, outcome assessment, and patient education. Current evidence is promising but heterogeneous. Image-based computer vision and deep learning appear closest to clinical use, whereas large language and generative image models require careful oversight. This review summarizes practical applications and proposes a cautious implementation pathway.","[""Shiraishi Makoto"", ""Okazaki Mutsumi""]",Oral and maxillofacial surgery clinics of North America,2026-10-14,10.1016/j.coms.2026.09.006,"[""Journal Article"", ""Review""]","[""Humans"", ""Artificial Intelligence"", ""Blepharoplasty"", ""Generative Artificial Intelligence"", ""Imaging, Three-Dimensional"", ""Deep Learning"", ""Patient Education as Topic""]", 42868900,https://pubmed.ncbi.nlm.nih.gov/42868900/,Maximizing mutual information in multi-view representation learning for next POI prediction.,"Next point-of-interest (POI) prediction is a fundamental task in spatiotemporal data mining, aiming to facilitate personalized location-based services through modeling individual mobility trajectories. Despite recent advances in deep neural networks for capturing human mobility patterns from both short- and long-term trajectories, accurate POI prediction remains challenging due to the fragility of short-term intent representations and user preference drift. These challenges arise from intent inconsistency between stable long-term behavioral patterns and transient short-term contingencies, as well as the inherent stochasticity induced by sparse, noisy, and exploratory human mobility trajectories, ultimately causing biased or unstable future intention predictions. In this study, we propose MVMI, a Multi-View representation learning framework based on Mutual Information for next POI prediction. To improve the robustness of short-term intent representations, MVMI decomposes short-term trajectories into instantaneous and context-dependent intents and employs a Gaussian perturbation-based multi-view augmentation strategy to enhance the robustness of short-term intent representations against trajectory stochasticity. To address user preference drift, MVMI further introduces a mutual information (MI) constraint that aligns integrated user preferences with the ground-truth future intentions, where a dual-task learning paradigm is employed to enforce hierarchical spatial constraints at both the POI and region levels. Finally, experiments on four real-world datasets demonstrate that MVMI significantly outperforms representative baselines.","[""Li Tuohang"", ""Yu Jiajing"", ""Zhao Jianyong"", ""Xie Yanzhe"", ""Huang Li"", ""Gao Qiang""]",Neural networks : the official journal of the International Neural Network Society,2026-10-07,10.1016/j.neunet.2026.109733,"[""Journal Article""]","[""Humans"", ""Representation Machine Learning"", ""Prediction Algorithms"", ""Predictive Learning Models"", ""Neural Networks, Computer"", ""Intention"", ""Data Mining""]", 42868899,https://pubmed.ncbi.nlm.nih.gov/42868899/,SegVLM: Deformable attentive visual enhancement for referring image segmentation using vision-language model.,"Image segmentation is a fundamental task in computer vision that involves partitioning an image into semantically meaningful regions. Referring image segmentation extends this task by using natural language expressions to localize specific objects, requiring the effective integration of visual and linguistic information. In this work, we propose SegVLM, a vision-language model that incorporates architectural enhancements to improve both segmentation accuracy and cross-modal alignment. The model integrates Squeeze-and-Excitation (SE) module for dynamic feature recalibration, deformable convolutions for geometric adaptability, and residual connection for deep feature learning. We also introduce a novel Referring-Aware Fusion (RAF) loss, which balances region-level alignment, boundary precision, and class imbalance. Extensive experiments and ablation studies demonstrate that each component contributes to consistent performance improvements. SegVLM further exhibits strong generalization across diverse datasets and a variety of referring expression scenarios.","[""Dalaq Alaa"", ""Behzad Muzammil""]",Neural networks : the official journal of the International Neural Network Society,2026-10-08,10.1016/j.neunet.2026.109721,"[""Journal Article""]","[""Image Processing, Computer-Assisted"", ""Humans"", ""Deep Learning"", ""Artificial Intelligence""]", 42863211,https://pubmed.ncbi.nlm.nih.gov/42863211/,Online continual Fréchet distance learning.,"Task-Free Continual Learning (TFCL) represents a realistic yet challenging learning scenario, aiming to train a model capable of capturing all previously seen important information without forgetting while the task information is not available. Existing studies usually propose learning a fixed-capacity model, leading to a suboptimal performance on complex data. In this research study, we propose a novel scalable mixture model framework to deal with both supervised and unsupervised learning scenarios under TFCL. A Fréchet distance-based model expansion mechanism is proposed to evaluate the knowledge discrepancy between each previously learned component of the mixture and that being currently trained, and use this measure for guiding model expansion. This mechanism is shown to promote the knowledge discrepancy among model's mixture components while enforcing the acquired information diversification, leading to a more compact model. We also propose a novel memory optimization mechanism that utilizes the entire previously learned knowledge to guide how to preserve novel data used for training a new mixture component, further promoting knowledge diversity among model's components. The empirical results, following a series of supervised and unsupervised continual learning experiments, show that the proposed framework achieves state-of-the-art performance.","[""Ye Fei"", ""Bors Adrian G""]",Neural networks : the official journal of the International Neural Network Society,2026-10-07,10.1016/j.neunet.2026.109728,"[""Journal Article""]","[""Unsupervised Machine Learning"", ""Humans"", ""Algorithms"", ""Supervised Machine Learning"", ""Neural Networks, Computer"", ""Learning"", ""Machine Learning"", ""Memory""]", 42863208,https://pubmed.ncbi.nlm.nih.gov/42863208/,Adaptive parameter manifold learning for low-dimensional neural network optimization.,"Deep neural networks have achieved remarkable success across diverse domains, yet their massive over-parameterization introduces significant optimization challenges, particularly susceptibility to overfitting and noise memorization. While traditional stochastic gradient descent treats all parameters equally in the native parameter space, recent evidence suggests that neural network optimization occurs within much lower-dimensional subspaces. Existing parameter subspace methods either rely on random projections that ignore parameter space structure or employ linear dimensionality reduction that cannot capture complex geometric relationships. Through empirical analysis of training trajectories, we demonstrate that neural network parameters evolve along continuous, low-dimensional non-linear manifolds whose intrinsic dimensionality is substantially lower than the original parameter space. We propose Adaptive Parameter Manifold Learning (APML), a novel optimization framework that leverages Local Tangent Space Alignment to discover the underlying manifold structure from training trajectories and optimize neural networks on the parameter manifolds. Our method dynamically adapts the manifold throughout training to accommodate the changing geometry of the loss landscape. Extensive experiments on noisy label learning tasks across vision and language domains, as well as federated learning scenarios, demonstrate that APML achieves superior robustness and communication efficiency compared to existing subspace optimization methods.","[""Du Zhekai"", ""Ao Wei"", ""Liu Jieyan"", ""Hui Bei"", ""Lu Ke""]",Neural networks : the official journal of the International Neural Network Society,2026-10-08,10.1016/j.neunet.2026.109725,"[""Journal Article""]","[""Neural Networks, Computer"", ""Federated Learning"", ""Algorithms"", ""Adaptive Algorithms"", ""Dimensionality Reduction"", ""Humans"", ""Machine Learning"", ""Soft Computing""]", 42855344,https://pubmed.ncbi.nlm.nih.gov/42855344/,Artificial Intelligence and Oral Potentially Malignant Disorders: Predicting Malignant Transformation and Recurrence.,"Oral potentially malignant disorders (OPMDs) are heterogeneous conditions with a significantly increased risk of developing oral squamous cell carcinoma. Traditional risk assessment methods are often subjective, largely unidimensional, and insufficient for individualized prediction of malignant transformation, treatment response, or disease recurrence. Artificial intelligence (AI) models developed with diverse patient data may improve OPMD risk assessment and provide more individualized decision support to clinicians. As such, this review discusses AI's role, available models, current evidence, and implementation challenges to applying this technology for OPMD risk assessment and management.","[""Adeoye John"", ""Su Yu-Xiong""]",Oral and maxillofacial surgery clinics of North America,2026-10-09,10.1016/j.coms.2026.09.001,"[""Journal Article"", ""Review""]","[""Humans"", ""Mouth Neoplasms"", ""Artificial Intelligence"", ""Cell Transformation, Neoplastic"", ""Neoplasm Recurrence, Local"", ""Risk Assessment"", ""Carcinoma, Squamous Cell"", ""Precancerous Conditions"", ""Predictive Learning Models"", ""Prediction Algorithms""]", 42855343,https://pubmed.ncbi.nlm.nih.gov/42855343/,Barriers to Clinical Integration of Artificial Intelligence in Computer-Assisted Dental Implant Surgery: A Critical Review.,"Translating artificial intelligence (AI) from virtual implant planning to dental implant surgical execution is obstructed by technical, clinical, and socioethical barriers. This critical review appraises these impediments, focusing on tomographic data heterogeneity, closed hardware ecosystems that block kinematic data extraction, and algorithmic opacity that erodes clinician trust. Current literature over-relies on static retrospective validations with high bias risk. This gap should be bridged by a roadmap prioritizing federated learning, open-source interoperability, explainable AI, and prospective multicenter trials to establish algorithms as dynamically adaptive surgical systems.","[""Chen Jinyan"", ""Wang Feng""]",Oral and maxillofacial surgery clinics of North America,2026-10-09,10.1016/j.coms.2026.09.008,"[""Journal Article"", ""Review""]","[""Humans"", ""Artificial Intelligence"", ""Surgery, Computer-Assisted"", ""Federated Learning"", ""Algorithms"", ""Dental Implantation, Endosseous"", ""Dental Implantation""]", 42854394,https://pubmed.ncbi.nlm.nih.gov/42854394/,Enhancing local robustness with Hölder regularization.,"Deep neural networks (DNNs) are vulnerable to small input perturbations, resulting in instability and adversarial susceptibility. While adversarial training methods such as PGD-AT and TRADES improve robustness by generating worst-case adversarial examples, they do not explicitly regulate the model's local smoothness. In this paper, we propose Hölder Regularization (HR), a novel method to enhance local robustness by constraining the model's output variations based on Hölder continuity. Unlike conventional adversarial defenses, HR directly controls the functional smoothness of the model by regulating the relationship between input distances and output variations. We provide theoretical insights showing that the proposed regularization directly tightens an upper bound on the adversarial risk, thereby improving robustness. We also demonstrate that HR can be effectively combined with adversarial training, offering a complementary approach to existing techniques. We further compare HR with representative robustness baselines, including adversarial training methods and spectral normalization, showing that HR achieves superior or competitive performance across different perturbation settings. Experiments on MNIST, CIFAR-10, BSEQ, and ISOLET show that our method consistently improves robustness under PGD attacks, Gaussian noise, ℓ∞-perturbations, and feature masking. Furthermore, the proposed approach remains computationally efficient, as the Hölder regularization term is approximated via mini-batch sampling, introducing only a modest overhead.","[""Li Haiyan"", ""Zhao Dazhi""]",Neural networks : the official journal of the International Neural Network Society,2026-10-03,10.1016/j.neunet.2026.109701,"[""Journal Article""]","[""Neural Networks, Computer"", ""Algorithms"", ""Deep Learning"", ""Humans""]", 42849131,https://pubmed.ncbi.nlm.nih.gov/42849131/,Uncertainty quantification for trustworthy deep learning: Methods and measures.,"The deployment of deep neural networks in safety-critical domains demands reliable estimates of predictive confidence, yet conventional architectures lack principled uncertainty quantification. This survey provides a structured, critical review of methods for Uncertainty Quantification (UQ) in deep learning, scoped to ensemble-based and approximate Bayesian approaches and the measures used to summarize their outputs. Relative to existing UQ surveys, our contribution is depth on efficient ensemble approximations and single-pass methods, and a unified treatment that separates the method producing a predictive distribution from the measure that summarizes its uncertainty. We organize methods into five families: Bayesian neural networks, Monte Carlo Dropout, deep ensembles, efficient ensemble approximations, and last-layer or single-pass approaches. We situate adjacent work on evidential and prior networks, conformal prediction, and post-hoc calibration, together with the decision-time tasks of out-of-distribution detection and selective prediction. For each, we examine theoretical motivation, implementation, empirical performance, and limitations. We then review ensemble diversity theory and uncertainty measures and their decompositions, contrasting the entropy decomposition with pairwise divergence measures, and consolidate evaluation methodology so that our qualitative comparisons share a common basis. We close with a brief treatment of uncertainty in large language models and open research directions, including efficient epistemic measures for classification, last-layer diversity, diversity and calibration under shift, and hybrid architectures.","[""Gillis H Martin"", ""Trappenberg Thomas""]",Neural networks : the official journal of the International Neural Network Society,2026-10-03,10.1016/j.neunet.2026.109699,"[""Journal Article"", ""Review""]","[""Deep Learning"", ""Uncertainty"", ""Bayes Theorem"", ""Monte Carlo Method"", ""Ensemble Learning"", ""Humans"", ""Neural Networks, Computer""]", 42843308,https://pubmed.ncbi.nlm.nih.gov/42843308/,Exploring running classifications using foot inclination angle and ground reaction force clustering methods: Do traditional foot strike categories capture meaningful patterns in ground reactions forces?,"BACKGROUND: Footstrike pattern has been a factor considered in various aspects of running injury and performance. Different features of the ground reaction forces are thought to vary across footstrike patterns. However, comprehensive analyses of ground reaction forces across these patterns have been limited and novel analysis techniques may now allow for additional insights into underlying biomechanical features not captured with traditional methods. RESEARCH QUESTION: Do kinematic footstrike classifications capture underlying ground reaction force patterns detected with clustering techniques? METHODS: 103 (47 male) collegiate cross-country runners completed overground running trials at their self-selected training pace. Ground reaction forces were collected over five trials of right foot contacts. 3D motion capture was utilized to assess foot inclination angle and classified as rearfoot strike (RFS), mid-foot strike (MFS), or forefoot strike (FFS). Principle components analysis and hierarchical agglomerative clustering (HAC) were used to identify clusters derived from ground reaction force features (i.e., loading rate and impact peak). Agreement between classification schemes (foot inclination angle vs. ground reaction force clusters) were then assessed. RESULTS: MFS was the most prevalent (n = 58, 56.3%), followed by RFS (n = 28, 27.2%), then FFS (n = 17, 16.5%). Five clusters were derived from HAC. RFS, MFS, and FFS patterns were distributed across each of the clusters derived from HAC. RFS and FFS patterns tended to cluster together whereas MFS patterns were distributed primarily across two clusters. Ground reaction force time series were distinct for each cluster with differences more apparent in the high frequency domain. SIGNIFICANCE: Ground reaction force patterns may have more intricate differences than kinematic classifications may be able to detect.","[""Vannatta C Nathan"", ""Tara Vivek Kumar"", ""Kernozek Thomas W""]",Gait & posture,2026-10-06,10.1016/j.gaitpost.2026.110661,"[""Journal Article""]","[""Humans"", ""Running"", ""Biomechanical Phenomena"", ""Foot"", ""Male"", ""Cluster Analysis"", ""Clustering Algorithms"", ""Young Adult"", ""Female"", ""Gait"", ""Motion Capture"", ""Principal Component Analysis""]", 42843264,https://pubmed.ncbi.nlm.nih.gov/42843264/,TIC: Co-optimization of training and inference for efficient image-text retrieval.,"Embedding-based image-text retrieval methods achieve efficient inference by independently encoding different modalities into a shared space. However, their performance is constrained by a dual limitation: fragile representations during training due to insufficient exploration of the semantic manifold, and biased similarity metrics during inference caused by the ""hubness"" problem. To address these challenges, we propose a Training-Inference Co-optimization (TIC) framework that systematically enhances both stages of the retrieval pipeline. During training, we introduce Embedding Space Self-Expansion (ES2)-a novel augmentation strategy that generates multiple dynamic views of image representations via random attention masking. This effectively transforms static sample points into local probabilistic spaces, enabling the model to explore broader embedding boundaries and learn more robust features. We further incorporate pre-trained GloVe semantic priors to initialize the text encoder, thereby providing linguistic knowledge that improves cross-modal alignment. During inference, we apply Dual Softmax Calibration-a zero-parameter post-processing technique that recalibrates the similarity matrix by incorporating global distribution information to mitigate hubness artifacts. Extensive experiments on the MS-COCO and Flickr30K benchmarks demonstrate the effectiveness of our approach. Our method achieves state-of-the-art performance among embedding-based techniques, significantly outperforming strong baseline models-even when utilizing a more lightweight text encoder.","[""You Zhongxiang"", ""Yang Shiyuan"", ""Li Zhixin""]",Neural networks : the official journal of the International Neural Network Society,2026-09-30,10.1016/j.neunet.2026.109696,"[""Journal Article""]","[""Algorithms"", ""Information Storage and Retrieval"", ""Humans"", ""Semantics"", ""Representation Machine Learning"", ""Neural Networks, Computer""]", 42843262,https://pubmed.ncbi.nlm.nih.gov/42843262/,New results of deterministic learning-based knowledge fusion.,"Identifying the unknown dynamics of dynamical systems from observed data is common in science and engineering. Recently, deterministic learning has the ability to achieve locally-accurate modeling of unknown dynamics via neural network. However, locally-accurate modeling only contains local, limited dynamics knowledge along the input trajectory data, making it difficult to adapt to tasks in complex environments. To obtain broader dynamics knowledge, deterministic learning-based knowledge fusion is studied in this paper, i.e., the dynamics knowledge modeled from each individual trajectory via deterministic learning is integrated into a fusion network. Specifically, the fusion problem is formulated as an optimization problem by minimizing a global discrepancy measure over learned local models, which naturally leads to a weighted least-squares formulation with a properly designed weighting matrix that characterizes the optimal fusion dynamics and improves the reliability of the resulting fusion model. Additionally, the fusion algorithm is extended to a distributed fusion algorithm. By using the zero-gradient-sum strategy, the network weight of each node is proven to converge exponentially to the global optimal solution relying only on the weight information of adjacent nodes. Numerical and real-world benchmark results demonstrate the effectiveness of the proposed fusion algorithms.","[""Zhang Jinyuan"", ""Wu Weiming"", ""Guo Junnan"", ""Hu Jingtao"", ""Duan Qiang"", ""Wang Cong""]",Neural networks : the official journal of the International Neural Network Society,2026-10-03,10.1016/j.neunet.2026.109703,"[""Journal Article""]","[""Neural Networks, Computer"", ""Algorithms"", ""Learning"", ""Soft Computing"", ""Machine Learning""]", 42837718,https://pubmed.ncbi.nlm.nih.gov/42837718/,Fusion-CapsNet: A transfer-driven capsule network trained on fused features and generative ai augmentation for strabismus detection.,"Strabismus is a common eye condition in which the eyes are does not line up correctly that impacts peoples of all ages. For effective treatment and to prevent durable problems, early detection and accurate classification are essential. This research presents a Transfer-driven Capsule Network trained on Fused features and generative AI augmentation method for strabismus detection, named as Fusion-CapsNet. Firstly, an improved Wasserstein Generative Adversarial Network with Gradient Penalty (ImWGAN-GP) method is utilized to generate high-quality synthetic eye images, which improves enhance the data diversity and model generalization. Following image augmentation, a Medav (median-average) based preprocessing is utilized to avoid the noise while preserving the critical structural as well as edge information. Then, accurate localization of visual regions is obtained by using an improved TransUNet (ImTransUNet) segmentation model. It integrates improved activation functions and an Atrous Spatial Pyramid Pooling-Progressive Enhancement Module (AS-IPEM) to effectively capture both local spatial details and global contextual information. From the segmented eye regions, multiple complementary features are extracted to widely characterize the strabismic patterns such as Eye Aspect Ratio (EAR), Local Gabor Transitional Patterns (LGTrP) and Active Shape Models (ASM). An improved weighted PCA with adaptive concatenation and dynamic scaling (ImWPCA-ACDS) based feature fusion strategy is introduced to effectively integrate these heterogeneous features. It preserves the semantic consistency, captures long-range dependencies, reduces feature dimensionality and mitigates overfitting. Lastly, strabismus detection is performed via an improved CapsNet with transfer learning (ImCapsNet-TL), which enables the model to effectively capture the hierarchical spatial relationships and long-range feature interactions while reducing the computational overhead. Experimental outcomes validate that the proposed Fusion-CapsNet framework obtained superior detection performance compared to standard methods. The ImCapsNet-TL detection frameworks has resulted higher accuracy by 97.7% in data size of 90%.","[""Gaikwad Priyanka Lahu"", ""Tambe Sagar Bhimraj""]",Computational biology and chemistry,2026-10-03,10.1016/j.compbiolchem.2026.109473,"[""Journal Article""]","[""Strabismus"", ""Humans"", ""Generative Artificial Intelligence"", ""Generative Adversarial Networks"", ""Algorithms"", ""Image Processing, Computer-Assisted""]", 42833029,https://pubmed.ncbi.nlm.nih.gov/42833029/,POTN-2: A deep learning-based predictor of HLA-A*02:01-restricted tumor neoantigens.,"Tumor neoantigens arise from somatic mutations in malignant cells and are strictly tumor-specific, minimizing autoimmune risks, making them ideal targets for T-cell immunotherapy. However, conventional pan-HLA predictors suffer from reduced accuracy and high false-positive rates for individual HLA subtypes. Our prior POTN model was trained on experimentally validated HLA-A02:01 peptide data. We developed POTN-2, an allele-specific predictor exclusively for HLA-A02:01-restricted tumor neoantigens, by expanding the feature repertoire and algorithmic architecture. We curated a comprehensive dataset comprising 50,308 experimentally confirmed positive binders, 117,334 randomly sampled negative controls, and 1012 matched mutant-wild-type neoantigen pairs. Multi-dimensional descriptors integrated, covering peptide length, residue positional encoding, amino acid composition, biological functional annotations, physicochemical properties. A three-stage feature filtering pipeline yielded three optimized subsets with 59, 65, and 109 features. We benchmarked Random Forest, Convolutional Neural Network (CNN), Support Vector Machine. The CNN trained on the full 109-feature subset achieved optimal performance, with classification accuracy of 0.936 across tumor antigens. This CNN-based POTN-2 achieved comparable performance to NetMHCpan-4.0 and DeepAntigen on tumor-associated antigen prediction, while substantially outperforming both tools in discriminating mutated neoantigen peptides from wild-type counterparts-a critical step for reducing false-positive candidates in personalized immunotherapy pipelines. We deployed an online platform (potn-2.cn)) that accepts FASTA, Excel, clinical sequencing inputs, balancing high computational throughput with subtype-specific precision to prioritize candidate neoantigens from mutation datasets. POTN-2 establishes a robust technical foundation for future extension to additional HLA alleles, and the platform adapts to emerging allele-specific models, facilitating personalized cancer immunotherapy research.","[""Song Linfeng"", ""Xu Chunrui"", ""Xiong Shiyi"", ""Zhang Junzhe"", ""Wang Yuran"", ""Wu Yahong"", ""Du Jiangfeng""]",Computational biology and chemistry,2026-09-29,10.1016/j.compbiolchem.2026.109460,"[""Journal Article""]","[""Humans"", ""Deep Learning"", ""HLA-A2 Antigen"", ""Antigens, Neoplasm"", ""Prediction Algorithms"", ""Neoplasms"", ""Immunoinformatics"", ""Predictive Learning Models"", ""Classification Algorithms"", ""Convolutional Neural Networks""]", 42833027,https://pubmed.ncbi.nlm.nih.gov/42833027/,"Integrated computational framework for drug repurposing in metastatic prostate cancer using transcriptomic meta-analysis, machine learning, and molecular modeling.","BACKGROUND: Metastatic prostate cancer (mPCa) continues to be the primary cause of mortality associated with prostate cancer, after considerable progress in targeted therapy. The rise of therapeutic resistance, coupled with the protracted and expensive nature of new drug development, has heightened interest in computational drug repurposing methodologies. This study sought to create a comprehensive computational framework that integrates transcriptome meta-analysis, network biology, machine learning, and molecular modelling to discover FDA-approved medications with potential therapeutic efficacy against metastatic prostate cancer. METHODS: Three publicly available Gene Expression Omnibus datasets were integrated following quality control, normalization. Differentially expressed genes between primary and metastatic prostate cancer samples were discovered and assessed for functional pathway enrichment and protein-protein interaction networks to identify hub proteins. A machine learning-based quantitative structure-activity relationship model was created and later utilized on FDA-approved pharmaceuticals. Candidate compounds were prioritized using molecular docking, ADMET prediction, and molecular dynamics simulations. RESULTS: A total of 236 DEGs were discovered, demonstrating considerable enrichment in pathways linked to cell cycle regulation and metastasis. CDK1 emerged as the hub gene. The optimized Light Gradient Boosting Machine model identified 67 FDA-approved compounds with predicted inhibitory activity against CDK1. Molecular docking and pharmacokinetic analyses prioritized Tucatinib and Sildenafil, and molecular dynamics simulations demonstrated superior structural stability and enhanced protein-ligand interactions for the CDK1-Tucatinib complex. CONCLUSION: This comprehensive computational framework offers a robust approach for therapeutic repurposing in metastatic prostate cancer. CDK1 was identified as a promising therapeutic target, and Tucatinib emerged as the most promising repurposing candidate, warranting further clinical validation.","[""Rafiee Javad"", ""Arabi Baygi Seyedeh Tahereh"", ""Aghaee-Bakhtiari Seyed Hamid""]",Computational biology and chemistry,2026-09-30,10.1016/j.compbiolchem.2026.109458,"[""Journal Article""]","[""Prostatic Neoplasms"", ""Male"", ""Drug Repositioning"", ""Machine Learning"", ""Humans"", ""Antineoplastic Agents"", ""Transcriptome"", ""Molecular Docking Simulation"", ""Molecular Dynamics Simulation"", ""Gene Expression Profiling"", ""Neoplasm Metastasis"", ""Quantitative Structure-Activity Relationship""]", 42832902,https://pubmed.ncbi.nlm.nih.gov/42832902/,Latent knowledge-driven unified representation learning for brain tumor segmentation with missing modalities.,"Gliomas are the most common primary malignant brain tumors, characterized by high heterogeneity, diffuse infiltration, and poor prognosis. Accurate tumor segmentation plays a fundamental role in diagnosis, treatment planning, and quantitative analysis. Although multimodal MRI provides rich anatomical and functional information, missing imaging modalities are frequently encountered in real-world clinical practice. Here, we propose a Latent Knowledge-driven Unified Representation Learning (LK-URL) framework for robust glioma segmentation under missing-modality conditions. LK-URL incorporates a Dynamic Adaptive Unified Interaction Module (DAUIM) to model intra- and inter-modality interactions, enabling enhanced tumor-aware feature representation and cross-modal compensation. Furthermore, a Latent Knowledge Collaborative Learning (LKCL) strategy is introduced to transfer structural knowledge from complete to incomplete modality settings via internal full-modality supervision, thereby improving the consistency of feature representations across diverse modality combinations and enhancing segmentation robustness. Extensive experiments on three datasets demonstrate that the proposed method consistently outperforms state-of-the-art approaches under missing-modality scenarios, while exhibiting strong generalization ability in real-world clinical settings. The code is available at https://github.com/joey-AI-medical-learning/LK-URL.","[""Ju Mengyi"", ""Gou Shuiping"", ""Jing Jingchen"", ""Fan Hongbin"", ""Li Yingping"", ""Zhang Chi""]",Neural networks : the official journal of the International Neural Network Society,2026-10-01,10.1016/j.neunet.2026.109695,"[""Journal Article""]","[""Brain Neoplasms"", ""Humans"", ""Representation Machine Learning"", ""Magnetic Resonance Imaging"", ""Glioma""]", 42828974,https://pubmed.ncbi.nlm.nih.gov/42828974/,Bulk and single-cell transcriptomics identify tumor-specific epithelial PSAT1 as a ferroptosis-associated therapeutic target in cervical cancer.,"Cervical cancer, the fourth most frequent malignancy in women worldwide, continues to be a major public health concern. Around 90% of high-grade cervical cancer cases are linked to infection with human papillomavirus (HPV). This phenotype, histologically, includes two major subtypes of cervical cancer: squamous cell carcinoma and adenocarcinoma. Drug resistance poses a major hurdle to treatment because it targets the apoptotic pathway. To overcome this, there is an urgent need for alternative routes to a curative strategy. Here, we focused on identifying non-apoptotic cell death pathways as a potential therapeutic option. Bulk RNA sequencing data for cervical cancer samples were obtained from the TCGA-CESC dataset. We combined gene set enrichment analysis, co-expression network analysis, and machine learning, which revealed ferroptosis as a key vulnerability at the bulk RNA level. Single-cell RNA sequencing data further confirmed the expression of ferroptosis-related components, including SCC and ADC. Subsequent analyses focused on identifying key regulators of the ferroptosis pathway. Further dissection of ferroptosis regulators and PSAT1 was prioritized as a prime translational target in adenocarcinoma compared with squamous cell carcinoma. To advance therapeutic translation, we applied AI-driven de novo compound generation to design a candidate drug targeting PSAT1. Additionally, molecular docking and molecular dynamics simulations identified GEN_42 as a putative PSAT1-binding compound. These findings nominate PSAT1 as a therapeutic vulnerability in cervical adenocarcinoma and GEN_42 as a computationally generated lead compound for further biological, cellular, and in vivo analyses to elucidate ferroptosis-induced cell death through modulation of PSAT1 function.","[""Tabassum Zainab"", ""Hadi Kabeer"", ""Sudheer Akkiraju"", ""Donakonda Sainitin""]",Computational biology and chemistry,2026-09-29,10.1016/j.compbiolchem.2026.109447,"[""Journal Article""]","[""Ferroptosis"", ""Female"", ""Humans"", ""Uterine Cervical Neoplasms"", ""Transcriptome"", ""Single-Cell Analysis""]", 42828972,https://pubmed.ncbi.nlm.nih.gov/42828972/,Cervical cancer survival analysis by integrating multi-omics data through an adaptive weighted deep neural network.,"Accurately analyzing cervical cancer prognosis holds significant importance for enhancing clinical outcomes within the framework of precision medicine. Since single-omics data only capture a limited perspective of cervical cancer patients, integrating multiple types of omics data is essential to provide more comprehensive insights. However, capturing the key information in multi-omics data fusion remains challenging owing to the presence of a large number of redundant variables and a relatively small sample size. In this study, we propose a deep neural network called AWMP to better integrate multi-omics data for cervical cancer survival analysis. The experiment demonstrated that AWMP enhances cancer prognosis prediction by 4.74% relative to the mean performance of the four comparator methods, benefiting from its adaptive weighted mechanism that effectively addresses the greedy nature of learning in multi-omics integration. In this study, based on the risk subgroups stratified by AWMP, four key genes (CDC42BPB, KLRG1, MAP7, LIMD2) are identified as been highly associated with cervical cancer prognosis and immune cell infiltration, highlighting their role in tumor heterogeneity. Additionally, single-cell tumor heterogeneity analysis revealed that the B-cell C2 subset correlates with both prognosis and immune response in cervical cancer, suggesting its potential as a therapeutic target. These findings underscore the accuracy and biological relevance of our deep learning framework for multi-omics integration in cervical cancer prognosis. Collectively, the identification of B-cell C2 subset advances the understanding of the precise role of B cells in cancer immune evasion and paves the way for more effective immunotherapy for cervical cancer.","[""Zheng Jinyong"", ""Su Ziyao"", ""Zhuo Xiaohui"", ""Chai Hua"", ""Tang Hui"", ""Lan Tianjun""]",Computational biology and chemistry,2026-10-01,10.1016/j.compbiolchem.2026.109465,"[""Journal Article""]","[""Humans"", ""Female"", ""Uterine Cervical Neoplasms"", ""Multiomics"", ""Neural Networks, Computer"", ""Survival Analysis"", ""Deep Learning"", ""Prognosis""]", 42826870,https://pubmed.ncbi.nlm.nih.gov/42826870/,Integrated transcriptomic and deep learning analyses define acRG signature for prognosis and therapeutic options in osteosarcoma.,"Osteosarcoma is a highly malignant bone tumor characterized by significant clinical heterogeneity and poor prognostic outcomes. In this study, we integrated single-cell RNA sequencing data to characterize the landscape of genes related to N4-acetylcytidine (ac4C) modification. We calculated an acRG activity score and found that activity was predominantly enriched in osteoblastic cells. Using a machine learning framework, we constructed a six-gene prognostic signature termed acRGBS. This signature served as a robust and independent predictor of overall survival across multiple independent cohorts. Further analysis revealed that high acRGBS scores were associated with an immunosuppressive tumor microenvironment and increased immune evasion potential. Among the hub genes identified, S100A13 was highlighted as a potential driver of tumor progression. Functional experiments demonstrated that S100A13 knockdown significantly inhibited cell proliferation and reduced xenograft tumor growth in vivo. Mechanistically, S100A13 physically interacts with the mitochondrial calcium uniporter (MCU) to maintain mitochondrial calcium homeostasis. Silencing S100A13 led to mitochondrial calcium overload, which exacerbated oxidative stress and induced cell apoptosis. Finally, we showed that S100A13 knockdown significantly enhanced the anti-tumor efficacy of cisplatin treatment. These findings suggest that S100A13 is a promising therapeutic target and a valuable biomarker for risk stratification in patients with osteosarcoma.","[""Xiao Honglei"", ""Yang Xiaolong"", ""Zhou Maobin"", ""Chang Hongze"", ""Zhang Yan"", ""Liu Xiaodong"", ""Cai Feng""]",Biochimica et biophysica acta. Molecular basis of disease,2026-10-02,10.1016/j.bbadis.2026.168492,"[""Journal Article""]","[""Osteosarcoma"", ""Humans"", ""Animals"", ""Prognosis"", ""Mice"", ""Bone Neoplasms"", ""Deep Learning"", ""Gene Expression Regulation, Neoplastic"", ""Transcriptome"", ""S100 Proteins"", ""Cell Line, Tumor"", ""Biomarkers, Tumor"", ""Cell Proliferation"", ""Apoptosis"", ""Female"", ""Tumor Microenvironment"", ""Xenograft Model Antitumor Assays"", ""Gene Expression Profiling""]", 42826577,https://pubmed.ncbi.nlm.nih.gov/42826577/,Machine learning-guided discovery of a novel natural CDK7 inhibitor for triple-negative breast cancer.,"Triple-negative breast cancer (TNBC) is an aggressive breast cancer subtype with limited targeted therapeutic options; while sacituzumab govitecan and PARP inhibitors (for BRCA-mutated TNBC) have received regulatory approval in specific patient subgroups, the majority of TNBC patients continue to lack effective targeted therapy. Cyclin-Dependent Kinase 7 (CDK7; UniProt P50613) is a crucial regulator of transcription and a promising therapeutic target. Natural compounds are a rich source of bioactive scaffolds, though they remain largely unexplored as CDK7 inhibitors. We applied a combined computational pipeline of machine learning (ML), virtual screening, molecular docking, molecular dynamics (MD) simulation, and MM-GBSA binding free energy analysis. We developed an ensemble ML model (LightGBM, XGBoost, CatBoost, and Random Forest; cross-validation R² = 0.7402) using a curated dataset of 935 CDK7 bioactivity compounds (spanning active through weakly active/inactive, pIC₅₀ range 3.18-9.51). The model evaluated 738,824 COCONUT natural compounds and identified 300 candidates for docking against CDK7 (PDB: 6XD3). The top five compounds underwent 10 ns MD simulations and MM-GBSA analysis, and the top hit was further examined with a 500 ns MD simulation. COC_081808, an indolocarbazole alkaloid (staurosporine family), emerged as the top candidate, exhibiting the best docking score (-12.28 kcal/mol), single-frame MM-GBSA (-34.18 kcal/mol), and trajectory MM-GBSA (-46.51 ± 3.82 kcal/mol). It formed stable hydrogen bonds with ASP155, ASN141, and ASN142 and π-π interactions with PHE91 and PHE93. Comparative MD showed the lowest ligand RMSD (1.01 ± 0.21 Å) and high hydrogen-bond occupancy. The 500 ns simulation verified stable protein-ligand interactions, with a protein RMSD of 1.87 Å, constant compactness, and the predominant low-energy conformational basin from PCA. COC_081808 is a computationally prioritized candidate with consistent multi-level evidence supporting potential CDK7 binding; experimental validation, including biochemical IC₅₀ determination, cellular assays, and kinome-wide selectivity profiling, remains essential before drawing therapeutic conclusions.","[""Kumar Chagaleti Bharath"", ""Pears Gandi Sony""]",Computational biology and chemistry,2026-10-01,10.1016/j.compbiolchem.2026.109466,"[""Journal Article""]","[""Humans"", ""Cyclin-Dependent Kinase-Activating Kinase"", ""Machine Learning"", ""Protein Kinase Inhibitors"", ""Triple Negative Breast Neoplasms"", ""Cyclin-Dependent Kinases"", ""Molecular Docking Simulation"", ""Molecular Dynamics Simulation"", ""Drug Discovery"", ""Biological Products"", ""Antineoplastic Agents"", ""Female""]", 42826576,https://pubmed.ncbi.nlm.nih.gov/42826576/,EXPSO: A PSO-based ensemble gene expression classifier for plant species using histone and chromatin accessiblity data.,"Gene expression in plants is regulated by complex interactions among chromatin features, including histone modifications and chromatin accessibility. Accurate prediction of gene-expression states from epigenomic information can provide insights into transcriptional regulation and facilitate functional genomic studies. However, the nonlinear and high-dimensional nature of epigenomic data presents challenges for conventional statistical approaches. Here, we developed EXPSO, a PSO-assisted ensemble learning framework for classifying plant genes into high- and low-expression states using integrated histone modification and chromatin accessibility data. The framework uses Particle Swarm Optimization (PSO) to optimize Random Forest hyperparameters within an ensemble comprising Random Forest, Support Vector Machine, and XGBoost classifiers. The framework was developed and evaluated independently across five plant species, including Arabidopsis thaliana, Hordeum vulgare, Phaseolus vulgaris, Setaria viridis, and Sorghum bicolor. EXPSO achieved an AUROC of 0.904 for Arabidopsis thaliana and demonstrated competitive predictive performance across the independently evaluated species. Model benchmarking against DeepChrome, ShallowChrome, and PatternChrome was performed using stratified five-fold cross-validation, with performance reported as mean ± standard deviation and 95% confidence intervals. Paired two-sided t-tests were used to assess fold-wise differences between EXPSO and the benchmark models. Comparative benchmarking included DeepChrome, ShallowChrome, and PatternChrome under the evaluation framework used in this study. Feature-attribution analyses using SHAP, LIME, and PDP consistently identified H3K36me3 and H3K56ac as important predictive features, consistent with their reported associations with transcription-associated chromatin states. To facilitate practical application, we developed the freely accessible EXPSO web server, which enables users to submit gene IDs together with histone modification and chromatin accessibility data as individual entries or CSV files and obtain gene-expression predictions with associated prediction probabilities and downloadable results. EXPSO provides an accessible and interpretable computational framework for plant gene-expression prediction from epigenomic features and a practical web-based implementation for supported plant datasets.","[""Sowmya H S"", ""Murmu Sneha"", ""Mishra Dwijesh Chandra"", ""Archak Sunil""]",Computational biology and chemistry,2026-09-21,10.1016/j.compbiolchem.2026.109426,"[""Journal Article""]","[""Chromatin"", ""Particle Swarm Optimization"", ""Histones"", ""Random Forest"", ""Arabidopsis"", ""Boosting Machine Learning Algorithms"", ""Plants"", ""Support Vector Machine"", ""Classification Algorithms"", ""Gene Expression Regulation, Plant""]", 42826575,https://pubmed.ncbi.nlm.nih.gov/42826575/,AI-driven transcriptomics for neurodegenerative disease research: A systematic review.,"BACKGROUND: Neurodegenerative diseases (NDDs), including Alzheimer's disease (AD), Parkinson's disease (PD), and Amyotrophic Lateral Sclerosis (ALS), present a growing global health challenge, with traditional diagnostic and therapeutic approaches facing significant limitations in early detection and mechanistic resolution. OBJECTIVES: Innovations in transcriptomic technologies-spanning bulk RNA sequencing, single-cell RNA-seq (scRNA-seq), and spatially resolved transcriptomics-provide unprecedented insights into cellular heterogeneity and microenvironmental dysregulation. This systematic review synthesizes recent applications of artificial intelligence (AI), machine learning (ML), and deep learning (DL) across transcriptomic tiers to benchmark computational models for biomarker identification, disease diagnosis, patient subtyping, and drug repurposing. METHODS & ELIGIBILITY: Following PRISMA 2020 guidelines for qualitative systematic reviews, literature published between 2016 and 2025 was systematically searched across PubMed/MEDLINE, IEEE Xplore, Google Scholar, and bioRxiv/medRxiv. Of 1058 records identified, 19 core empirical studies met all eligibility criteria and were synthesized in depth alongside contextual reference benchmarks. SYNTHESIS & KEY FINDINGS: Traditional ML algorithms (SVM, RF, Elastic Net) demonstrate high diagnostic accuracy (AUC 0.72-0.98) in blood-based biomarker selection, while deep architectures (autoencoders, LSTMs, and Graph Neural Networks) excel at modeling continuous disease trajectories and resolving spatial transcriptomic niches (e.g., STAGATE, SpaGCN, cell2location). Emerging spatial deep learning models successfully localize disease-associated microglia (DAM) around amyloid plaques and trace axonal degeneration pathways. LIMITATIONS & CONCLUSIONS: Key translational hurdles include data scarcity in rare NDD subtypes, lack of cross-platform spatial benchmarks, and the black-box nature of deep neural networks. Integrating explainable AI (XAI) and prospective multi-modal cohorts is essential for translating computational transcriptomics into clinical diagnostics and targeted therapeutics.","[""Tripathi Anushree"", ""Yadav Anoop Kumar"", ""Kumar Nilesh""]",Computational biology and chemistry,2026-09-29,10.1016/j.compbiolchem.2026.109448,"[""Journal Article"", ""Systematic Review"", ""Review""]","[""Humans"", ""Neurodegenerative Diseases"", ""Artificial Intelligence"", ""Transcriptome"", ""Deep Learning"", ""Machine Learning"", ""Biomarkers"", ""Gene Expression Profiling""]", 42826573,https://pubmed.ncbi.nlm.nih.gov/42826573/,FKAN-MoCBDTI: Multimodal drug-target interaction prediction via collaborative bilinear attention and knowledge distillation.,"Drug-target interaction (DTI) prediction plays a pivotal role in drug discovery and drug repurposing, yet current deep learning approaches still suffer from limitations in multimodal fusion, fine-grained cross-modal interaction modeling, nonlinear representation learning, and inductive generalization. To overcome these issues, we develop FKAN-MoCBDTI, a multimodal framework that unifies heterogeneous feature extraction, collaborative bilinear attention, Fourier Kolmogorov-Arnold Networks (FKAN), teacher-student knowledge distillation and conditional domain adversarial learning (CDAN). On the drug side, SMILES sequences are encoded by a Transformer and molecular topology is modeled by an FKAN-based graph neural network. On the target side, M-gram embeddings are combined with multi-layer graph convolutional networks to capture complementary sequential and sequence-derived topological information. A collaborative bilinear attention module is designed to explicitly characterize local pairwise associations between drug molecular and protein residue features, improving both feature fusion and qualitative attention visualization, while FKAN is used as the predictor to enhance high-order nonlinear modeling. Furthermore, we integrate a teacher-student KD mechanism with a CDAN module to effectively bridge domain discrepancies under cross-domain scenarios. Comprehensive evaluations across six benchmark datasets demonstrate the competitive performance of FKAN-MoCBDTI. Specifically, standard pair-wise random splitting validates in-domain pair interpolation, while evaluations across cold-entity settings (cold-drug and cold-both) and clustering-based cross-domain transfers demonstrate its strong attribute-based inductive generalization on unseen chemical compounds and novel interaction pairs, alongside nuanced performance trade-offs under unseen target protein (cold-target) scenarios. Ablation studies, visualization analyses, and retrospective case studies further confirm the effectiveness of the proposed framework, providing structural and literature-based support for the plausibility of selected predictions.Overall, FKAN-MoCBDTI offers a robust and generalizable approach for DTI prediction.","[""Hu Xianjun"", ""Zhang Zhen"", ""Deng Ziyan"", ""Wang Lei""]",Computational biology and chemistry,2026-09-29,10.1016/j.compbiolchem.2026.109450,"[""Journal Article""]","[""Pharmaceutical Preparations"", ""Drug Discovery"", ""Graph Neural Networks"", ""Deep Learning"", ""Proteins""]", 42826571,https://pubmed.ncbi.nlm.nih.gov/42826571/,The benefits and biases of artificial intelligence in microbiology.,"Microbiology has been greatly advanced by the incorporation of Artificial Intelligence (AI) by improving the classification, isolation and detection of microorganisms in several fields where microbial exploration is carried out. With the ability to process large volumes of data from genomic sequences, biochemical profiles and imaging, AI enables faster and more accurate microbial identification than conventional techniques. In microbiome study, AI decodes complex microbial communities in environments, the human body and air by identifying unknown species and their ecological roles. AI is a valuable tool in predicting antimicrobial resistance (AMR) by detecting resistance genes within microbial genomes and aiding in effective treatment strategies. AI-powered diagnostic tools and biosensors offer rapid pathogen detection, while Natural Language Processing (NLP) aids in tracking emerging microbial threats through scientific literature and health databases. In drug and vaccine development, AI accelerates the discovery process by simulating molecular interactions and predicting outcomes, saving time and resources. Most AI models, especially deep learning tools, however, lack transparency leading to difficult result interpretation. Additionally, over-fitting and limited generalizability of AI models are also concerns, along with ethical and legal issues like data privacy in microbiome research and clinical settings. Prior review articles have examined AI's impact in singular microbiological settings, and special attention has been paid to AI applications in medical microbiology. This review article reaches out and analyzes AI applications across general, industrial, medical, and environmental microbiology. It also outlines how mitigation strategies like inclusive dataset design, algorithm vigilance, evaluation techniques like Translational Evaluation of Healthcare AI (TEHAI) and emerging directions of analyses like federated learning and Explainable Artificial Intelligence (XAI) address the risks associated with AI use like bias and reduced interpretability.","[""Aransiola Sesan Abiodun"", ""Onyinoyi Adams Husseinat"", ""Akhigbe Joshua O"", ""Onoruoiza Suleiman Ishaq"", ""Esenogho Ebenezer"", ""Maddela Naga Raju""]",Computational biology and chemistry,2026-09-30,10.1016/j.compbiolchem.2026.109457,"[""Journal Article"", ""Review""]","[""Artificial Intelligence"", ""Humans"", ""Microbiology"", ""Microbiota"", ""Deep Learning""]", 42826523,https://pubmed.ncbi.nlm.nih.gov/42826523/,Reinforcement learning-based predefined-time secure control for fractional-order nonlinear systems with sensor and actuator attacks.,"A secure control problem is investigated for constrained fractional-order nonlinear systems subject to both sensor and actuator attacks. To handle the effects induced by the dual attacks, a nonlinear mapping-based coordinate transformation is elaborately introduced to convert the original state-constrained system into an unconstrained form. Furthermore, by integrating reinforcement learning with predefined-time stability analysis, an adaptive control strategy is developed to achieve predefined-time convergence. The proposed method enhances the transient response of the system while accelerating the convergence of both the tracking error and the policy learning process. A novel reset-based self-triggered mechanism is proposed. On the one hand, the developed mechanism reduces communication load and eliminates the need for continuous state monitoring. On the other hand, the sampling intervals can be adaptively adjusted according to the system states, thereby further alleviating the computational burden. Under the developed control strategy, the closed-loop system is guaranteed to achieve semi-global practical predefined-time stability, and all signals converge to a bounded neighborhood of the origin within the predefined time.","[""Liang Chunyu"", ""Wang Fang"", ""Wang Zhen"", ""Lai Guanyu""]",Neural networks : the official journal of the International Neural Network Society,2026-09-29,10.1016/j.neunet.2026.109697,"[""Journal Article""]","[""Reinforcement Machine Learning"", ""Nonlinear Dynamics"", ""Algorithms"", ""Computer Security"", ""Soft Computing""]", 42822069,https://pubmed.ncbi.nlm.nih.gov/42822069/,L2AF: Graph-attention modulated learning automata for efficient multi-agent cooperation.,"Multi-agent reinforcement learning (MARL) is impeded by the combinatorial explosion of joint action spaces and the neglect of complex inter-agent dependencies, which severely hinder effective coordination and accurate credit as signment. To address these challenges, this paper proposes a novel framework termed Lightweight Learning Automata with Feature extraction (L2AF), integrated with a Graph-Attention Modulated Feature Extractor (GAMFE). Specifically, GAMFE synergizes compact MLP embeddings with graph attention networks to explicitly model relational dependencies among agents. This mechanism generates attention-weighted continuous reward signals (S-model) that dynamically modulate the update step size of the automata. Theoretically, via ordinary differential equation (ODE) modeling and Lyapunov stability analysis, we rigorously establish that L2AF achieves local asymptotic stability and converges to the global optimal joint action at an exponential rate. Experimental evaluations on stochastic coordination environments demonstrate that L2AF achieves an AUC of 0.77 ± 0.11, significantly outperforming state-of-the-art baselines such as LA-OCA and EMAQ. Furthermore, ablation studies confirm that the attention mechanism and continuous feedback yield performance gains of 6.9% and 11.6%, respectively, while their integration reduces learning variance by 15%, highlighting the framework's robustness.","[""Wang Jun"", ""Ke Wenjie"", ""Liu Lu"", ""Dong Que"", ""Liu Jia"", ""Lai Pan"", ""Zhang Xiao""]",Neural networks : the official journal of the International Neural Network Society,2026-09-22,10.1016/j.neunet.2026.109676,"[""Journal Article""]","[""Reinforcement Machine Learning"", ""Attention"", ""Humans"", ""Graph Neural Networks"", ""Algorithms""]", 42809960,https://pubmed.ncbi.nlm.nih.gov/42809960/,Spectral-temporal dual view graph convolutional networks: A deep learning framework for dynamic bipartite graphs.,"Dynamic bipartite graphs (DBGs) are widely used in real-world scenarios, where representation learning is particularly challenging due to the heterogeneity of node types and the temporal evolution of interactions. A key difficulty lies in jointly capturing non-stationary micro-level preference dynamics and macro-level stable structural evolution. Existing methods typically focus on either temporal dependency modeling or structural pattern extraction. For instance, recurrent or attention-based temporal models struggle to capture abrupt, short-term preference shifts, while spectral or structure-oriented approaches emphasize long-term regularities but overlook fine-grained dynamic variations. This imbalance becomes especially pronounced in scenarios with sparse interactions or sudden behavioral changes. To address these challenges, this paper proposes a spectral-temporal dual-view framework, ST-DYB, for DBG representation learning. By explicitly decoupling macro-level structural evolution and micro-level preference dynamics, the model captures complementary heterogeneous patterns from two perspectives. A cross-view adaptive fusion mechanism further integrates these representations, enabling effective modeling of both implicit preference dependencies and temporal dynamics. Extensive experiments on multiple real-world datasets demonstrate that ST-DYB consistently outperforms state-of-the-art methods in dynamic link prediction and node classification tasks, validating the effectiveness and robustness of the proposed dual-view modeling paradigm.","[""Xing Zhezhe"", ""Ye Yuxin"", ""Li Ziheng"", ""Liu Jiawen""]",Neural networks : the official journal of the International Neural Network Society,2026-09-25,10.1016/j.neunet.2026.109672,"[""Journal Article""]","[""Deep Learning"", ""Graph Neural Networks"", ""Representation Machine Learning"", ""Humans"", ""Algorithms""]", 42809958,https://pubmed.ncbi.nlm.nih.gov/42809958/,Cross-domain few-shot learning for hyperspectral image classification based on mixup foundation model.,"Although cross-domain few-shot learning (CDFSL) for hyperspectral image (HSI) classification has attracted significant research interest, existing works often rely on an unrealistic data augmentation procedure in the form of external noise to enlarge the sample size, thus greatly simplifying the issue of data scarcity. They involve a large number of parameters for model updates, being prone to the overfitting problem. To the best of our knowledge, only one work has explored the strength of the foundation model, yet it still relies on a conventional foundation model for RGB images. This paper proposes the MIxup FOundation MOdel (MIFOMO) for CDFSL of HSI classifications. MIFOMO is built upon the concept of a remote sensing (RS) foundation model, pre-trained across a large scale of RS problems, thus featuring generalizable features. The notion of coalescent projection (CP) is introduced to quickly adapt the foundation model to downstream tasks while freezing the backbone network. The concept of mixup domain adaptation (MDM) is proposed to address the extreme domain discrepancy problem. Last but not least, the label propagation concept is implemented to cope with noisy pseudo-label problems. Our rigorous experiments demonstrate the advantage of MIFOMO, where it beats prior art by up to a 14% margin. The source code of MIFOMO is open-sourced in https://github.com/Naeem-Paeedeh/MIFOMO for reproducibility and further study.","[""Paeedeh Naeem"", ""Pratama Mahardhika"", ""Shiddiqi Ary"", ""Cao Zehong"", ""Prasad Mukesh"", ""Jatmiko Wisnu""]",Neural networks : the official journal of the International Neural Network Society,2026-09-25,10.1016/j.neunet.2026.109684,"[""Journal Article""]","[""Hyperspectral Imaging"", ""Neural Networks, Computer"", ""Machine Learning"", ""Humans"", ""Remote Sensing Technology"", ""Algorithms""]", 42809955,https://pubmed.ncbi.nlm.nih.gov/42809955/,Continual learning based state-aware estimation for reliable industrial legged robots.,"Reliable state estimation is fundamental to learning-based control of legged robots. Existing learning-based locomotion estimators typically predict only limited privileged states or learn implicit latent representations, while insufficiently addressing the non-stationary data distribution induced by evolving reinforcement-learning policies. In this setting, explicit multi-state estimation involves three coupled challenges: scalable estimation of heterogeneous robot states under onboard constraints, task-aware utilization of estimated states, and preservation of estimator plasticity during non-stationary training. To address these challenges, we propose SENSE, a state-aware estimation framework for reinforcement-learning-based legged locomotion. SENSE employs a sparse Mixture-of-Experts State Estimation Module (SEM) to infer diverse explicit robot states from proprioceptive histories, and introduces a StateSelector to forward task-relevant estimates to the control policy. We further propose CBP(COMP), a competitive continual backpropagation mechanism that replaces random neuron reinitialization with competitive parameter inheritance to alleviate unbalanced neuron turnover and preserve plasticity. Mechanism-level analysis explains its asymptotic competition symmetry under simplified assumptions. Controlled estimator comparisons, ablations, reset-distribution statistics, dynamic-payload adaptation, and real-world hexapod experiments show improved estimation accuracy, control performance, continual adaptation, and deployment robustness. These results suggest a practical neural state-estimation paradigm for interpretable, adaptive, and reliable learning-based legged robots.","[""Yang Xuekang"", ""Li Haoyang"", ""Zhu Jialing"", ""Ye Heng"", ""Sun Qiao"", ""Zhong Jiangwei"", ""Gao Yue""]",Neural networks : the official journal of the International Neural Network Society,2026-09-24,10.1016/j.neunet.2026.109686,"[""Journal Article""]","[""Robotics"", ""Reinforcement Machine Learning"", ""Humans"", ""Soft Computing"", ""Algorithms"", ""Neural Networks, Computer"", ""Autonomous Robots"", ""Adaptive Algorithms"", ""Industry""]", 42809954,https://pubmed.ncbi.nlm.nih.gov/42809954/,Transfer learning-based safe Q-learning for discrete-time nonlinear systems with asymmetric state constraints.,"This article investigates the problem of safe control for discrete-time (DT) nonlinear systems subject to asymmetric state constraints. A novel transfer learning-based safe Q-learning (TL-SQL) algorithm is developed within the adaptive dynamic programming (ADP) framework. To support the proposed TL-SQL algorithm, a novel revived transformation formulation for DT systems is derived, serving as the basis for transferring control policies from unconstrained source systems to constrained target systems. Building upon this foundation, the proposed TL-SQL algorithm improves policy transfer under asymmetric constraints in three aspects. Firstly, unlike traditional state transformation methods, the revived transformation alleviates the adverse impact of state transformation on the optimality of the original system. Secondly, compared with control-barrier-function-based approaches, it avoids complicating the utility function structure. Thirdly, by decoupling policy training from constraint handling, it enables efficient adaptation to changing constraints without retraining. Theoretical analysis ensures stability and safety of the transferred policy. Simulation results validate the effectiveness of TL-SQL under asymmetric constraint scenarios.","[""Chen Jiaoyuan"", ""Zhu Minglei"", ""Song Shijie"", ""Zhao Yuyang"", ""Gong Dawei""]",Neural networks : the official journal of the International Neural Network Society,2026-09-25,10.1016/j.neunet.2026.109685,"[""Journal Article""]","[""Nonlinear Dynamics"", ""Dynamic Programming"", ""Algorithms"", ""Neural Networks, Computer"", ""Computer Simulation"", ""Adaptive Algorithms"", ""Reinforcement Machine Learning"", ""Humans""]", 42805153,https://pubmed.ncbi.nlm.nih.gov/42805153/,Machine learning-assisted optimization of injectable silk fibroin hydrogels functionalized with decellularized extracellular matrix and mesenchymal stromal cell secretome.,"Silk fibroin (SF) hydrogels are promising injectable scaffolds for wound repair, but optimizing their multi-parametric gelation remains challenging due to complex formulation-processing interactions. We developed a bioinspired gelation strategy combining potassium-driven β-sheet nucleation with sonication-mediated conformational activation, guided by factorial design and machine learning. Artificial neural network (NN) and Random Forest (RF) algorithms mapped non-linear relationships between SF concentration, potassium (K+) concentration, sonication time, and gelation kinetics. Machine-learning-assisted analysis delineated an experimentally informed operational region associated with complete or near-complete gelation and generally favorable gelation kinetics, which was subsequently used to select formulations for technological characterization. All formulations showed physiological pH, acceptable injectability (extrusion forces <5 N), and sonication-tunable microstructure ranging from compact lamellar to porous sponge-like architectures. To enhance biological functionality, gels were combined with adipose-derived mesenchymal stromal cell-lyosecretome and human skin-derived decellularized extracellular matrix (dECM), with component-level testing supporting their cytocompatibility and identifying blood-interaction profiles that warrant further evaluation in the complete composite formulation. Functional analyses revealed distinct in vitro biological profiles: dECM produced the strongest pro-angiogenic response in the tube formation assay, accompanied by VEGFA upregulation and downregulation of NFKB1, HMGB1, and BAX in fibroblasts. Lyosecretome showed complementary effects, including selective SOD2 upregulation, FGF7 induction, and enhanced fibroblast migration when tested as a soluble component. SF-containing formulations attenuated fibroblast scratch closure relative to controls, indicating a matrix-dependent modulation of cell migration. These in vitro findings define distinct component-associated biological response profiles and provide a basis for subsequent mechanistic and preclinical evaluation.","[""Bertania Edoardo"", ""Modena Angelo"", ""Rinaldi Maurizio"", ""Tosca Marta Cecilia"", ""Miletto Ivana"", ""Lim Dmitry"", ""Renzulli Ilenia Rita"", ""Balderacchi Alice Maria"", ""Ferrarotti Ilaria"", ""Corsico Angelo Guido"", ""Bari Elia"", ""Sorlini Marzio"", ""Sesana Giovanni"", ""Torre Maria Luisa""]",Biomaterials advances,2026-09-21,10.1016/j.bioadv.2026.215193,"[""Journal Article""]","[""Mesenchymal Stem Cells"", ""Fibroins"", ""Humans"", ""Hydrogels"", ""Machine Learning"", ""Decellularized Extracellular Matrix"", ""Animals"", ""Tissue Scaffolds""]", 42805098,https://pubmed.ncbi.nlm.nih.gov/42805098/,Integrating hypergraph representation learning and gradient-boosted classification for multi-omics glioblastoma gene prioritization: A multi-stage learning framework.,"Cancer continues to be a significant worldwide health concern, mostly resulting from modifications in genetic material at the molecular level. The research introduces a hybrid system that integrates raw gene-level multi-omics features with pathway-derived hypergraph embeddings to prioritize glioblastoma genes. The proposed pipeline aggregates gene-level expression, mutation, and copy-number features; constructs a gene-pathway hypergraph from curated pathway memberships; and trains an attention-based hypergraph encoder (HAGNet) to produce structure-aware gene embeddings, evaluated by an XGBoost classifier. We assess the framework using stratified 5-fold cross-validation, reporting ROC-AUC, PR-AUC, Precision@K, and Recall@K at K=50, 100, and 300. Using only raw multi-omics features, the model achieved an ROC-AUC of 0.611 ± 0.020 and a PR-AUC of 0.095 ± 0.017, whereas HAGNet embeddings improved these metrics to 0.768 ± 0.021 and 0.202 ± 0.021, respectively. The integrated model achieved superior overall performance, with ROC-AUC of 0.777 ± 0.021 and PR-AUC of 0.209 ± 0.030, demonstrating that raw omics signals and hypergraph-based structural representations are synergistic. Among the evaluated classifiers, HAGNet + XGBoost achieved the strongest overall discrimination, with ROC-AUC and PR-AUC of 0.7833 ± 0.0102 and 0.2180 ± 0.0122, respectively, outperforming Random Forest, SVM, and Decision Tree. We additionally delineate robustness, ranking, pathway enrichment, and external validation analyses to substantiate biological interpretation. The findings indicate that pathway-aware hypergraph representation learning is a practical and effective enhancement to traditional gene-level features for TCGA-GBM gene prioritization.","[""Nowsheena"", ""Qureshi Shaima""]",Computational biology and chemistry,2026-09-28,10.1016/j.compbiolchem.2026.109441,"[""Journal Article""]","[""Glioblastoma"", ""Humans"", ""Boosting Machine Learning Algorithms"", ""Multiomics"", ""Representation Machine Learning"", ""Brain Neoplasms"", ""Classification Algorithms""]", 42805095,https://pubmed.ncbi.nlm.nih.gov/42805095/,"AI-driven drug repurposing: Methods, platforms, and translational applications.","The conventional drug discovery pipeline is resource-intensive, time-consuming, and characterized by high attrition rates, often requiring more than a decade and billions of dollars to bring a single therapeutic agent to market. Drug repurposing identifying new therapeutic indications for approved or investigational drugs offers a pragmatic alternative by leveraging established safety profiles and known pharmacokinetics. Despite growing interest, comprehensive syntheses that span the full spectrum of artificial intelligence (AI) methodologies from classical machine learning (ML) through deep learning (DL), graph neural networks (GNNs), and generative AI while also critically evaluating platforms, validation status, and translational readiness remain limited. This review addresses that gap. A narrative review methodology was employed. Literature was searched across PubMed, Scopus, Web of Science, and Google Scholar using a structured keyword strategy (terms: drug repurposing, AI drug discovery, machine learning, deep learning, graph neural networks, generative AI, reinforcement learning, clinical translation combined with Boolean operators). Studies published between 2018 and 2025 were prioritized, with foundational works included regardless of date. Inclusion criteria encompassed primary research articles, systematic reviews, and platform descriptions reporting AI-based drug repurposing methods or outcomes. Non-English publications, conference abstracts without full text, and purely theoretical works without empirical validation were excluded. We identified and synthesized evidence across six thematic domains: (1) classical ML strategies, including target-based and phenotypic screening; (2) DL architectures, including convolutional and recurrent networks; (3) GNNs for biological network modeling; (4) generative AI and reinforcement learning for de novo molecular design; (5) disease-specific applications in oncology, neurology, and COVID-19; and (6) bioinformatics databases underpinning model development. We explicitly compared ML, DL, GNN, and generative AI approaches in terms of data requirements, interpretability, scalability, and translational readiness and summarized these dimensions in a comparative table. Key platforms reviewed RepurposeDrugs, NeuroCADR, and COVID-19 repurposing frameworks were selected using predefined criteria that emphasized methodological transparency, accessibility, and reported validation. Across approaches, AI models frequently demonstrate strong internal predictive performance in silico, with graph-based and transformer models representing the current state of the art on benchmark datasets. However, external validation, reproducibility, and successful clinical implementation remain comparatively scarce. We therefore place particular emphasis on external cohort validation, prospective and experimental follow-up, and the distinction between computational promise and realized clinical benefit. Persistent challenges include dataset bias, limited model interpretability, regulatory uncertainty, and infrastructure gaps in low-resource settings. AI-driven drug repurposing holds considerable translational promise, but many claims remain preliminary. Interdisciplinary collaboration, shared benchmarks, transparent reporting of negative results, and prospective clinical validation frameworks are essential for realizing the full therapeutic potential of these approaches.","[""Malik Zeeshan Khalid"", ""Ahmad Kamran"", ""Ahmed Sheheryar""]",Computational biology and chemistry,2026-09-26,10.1016/j.compbiolchem.2026.109444,"[""Journal Article"", ""Review""]","[""Drug Repositioning"", ""Humans"", ""Artificial Intelligence"", ""Graph Neural Networks"", ""Generative Artificial Intelligence"", ""Deep Learning"", ""Drug Discovery"", ""Machine Learning"", ""Translational Research, Biomedical""]", 42801912,https://pubmed.ncbi.nlm.nih.gov/42801912/,Integrative analysis of single-cell and bulk RNA sequencing reveals inflammatory mechanisms of cancer-induced nerve injury and enables prognostic modeling in lung adenocarcinoma.,"BACKGROUND: Cancer-induced nerve injury (CINI) is implicated in unfavorable clinical outcomes across multiple cancer types. However, its biological role and prognostic value remain insufficiently characterized in lung adenocarcinoma (LUAD). METHODS: Bulk RNA sequencing, single-cell RNA sequencing, and spatial transcriptomics datasets were integrated to investigate CINI-related molecular features in LUAD. Survival analysis was performed to evaluate the prognostic significance of CINI-related transcriptional activity. Immune infiltration analysis, functional enrichment analysis, and tumor microenvironment (TME) profiling were conducted to explore the association between CINI and immune environment. Weighted gene co-expression network analysis (WGCNA) and machine learning algorithms were further applied to construct a CINI-related prognostic signature. Single-cell and spatial transcriptomic analyses were used to identify CINI-associated epithelial cells, characterize their developmental states, and explore potential intercellular communication patterns. In vitro experiments were performed to validate the functional relevance of the selected risk gene. RESULTS: Survival analysis indicated a significant negative association between CINI activity and overall survival in LUAD patients across multiple transcriptomic cohorts. Single-cell and spatial transcriptomic analyses identified malignant transcriptional features, increased developmental plasticity, and preferential enrichment in metastatic lesions of CINI-associated epithelial cells. We further established a CINI-related prognostic model, and in vitro experiments supported the functional relevance of PLEK2 in LUAD cell proliferation and migration. CONCLUSION: This study provided an integrative multi-omics characterization of CINI in LUAD and suggested that CINI activity was associated with TME remodeling and prognosis. These findings highlight the potential value of CINI-related signatures for prognostic stratification and future therapeutic exploration.","[""Wu Wentian"", ""Ji Pengcheng"", ""Zhang Ziyu"", ""Chen Sheng"", ""Zhao Mei"", ""Li Xinying"", ""Zhang Yiruo"", ""Du Yingying""]",Computational biology and chemistry,2026-09-24,10.1016/j.compbiolchem.2026.109433,"[""Journal Article""]","[""Humans"", ""Adenocarcinoma of Lung"", ""Prognosis"", ""Lung Neoplasms"", ""Sequence Analysis, RNA"", ""Single-Cell Analysis"", ""Tumor Microenvironment"", ""Inflammation"", ""Single-Cell Gene Expression Analysis"", ""Spatial Transcriptomics""]", 42800395,https://pubmed.ncbi.nlm.nih.gov/42800395/,DK-FocalNeXt: Deep Kronecker based FocalNeXt for skin cancer classification using dermoscopic images.,"The global burden of skin cancer continues to rise, posing a serious public health challenge. Accurate recognition is vital to ensure effective therapy and increase survival chances, even though classical diagnostic methods depend largely on the availability and experience of dermatologists. In many regions, especially those with limited medical infrastructure, it can lead to late diagnosis and negative impacts on patient health. Consequently, there is increasing interest in creating automated, precise, and accessible diagnostic tools to aid clinicians and facilitate early diagnosis of skin cancer. To moderate these complications, this article proposes an innovative approach named Deep Kronecker based FocalNeXt (DK-FocalNeXt) for the classification of skin cancer. The proposed framework initially takes skin images from the datasets as input. Then, image denoising is done using the Rudin-Osher-Fatemi model (ROF). The next stage involves segmenting the skin lesions using DAMMD-Net. Moreover, LogicMix is used for image augmentation, which promotes skin cancer classification by enhancing the diversity and size of training datasets. Thereafter, feature extraction was done to gain meaningful and distinctive representations of the lesions, which were essential for accurate classification. In the final stage, the skin lesions are classified into their respective categories using the DK-FocalNeXt, which is developed by integrating Deep Kronecker Neural Network (DKN) and FocalNeXt. Furthermore, DK-FocalNeXt has gained an accuracy of 96.61%, a weighted average precision of 95.69%, and a macro average recall of 95.38%.","[""Patil Uma"", ""Angadi Sanjeevkumar"", ""Hegde Nagaratna P"", ""Kale Bharati R""]",Computational biology and chemistry,2026-09-21,10.1016/j.compbiolchem.2026.109436,"[""Journal Article""]","[""Humans"", ""Skin Neoplasms"", ""Dermoscopy"", ""Image Processing, Computer-Assisted"", ""Neural Networks, Computer""]", 42800394,https://pubmed.ncbi.nlm.nih.gov/42800394/,In silico identification of robust stage-specific gene signatures in neuroblastoma: Integrating microarray meta-analysis with machine learning and cross-platform validation.,"Neuroblastoma (NB) exhibits profound clinical and molecular heterogeneity, highlighting the need for reproducible biomarkers for disease stratification. This study aimed to identify robust stage-associated gene signatures through integrative machine learning (ML) and cross-platform validation. Four independent NB microarray datasets were integrated to perform a multi-cohort meta-analysis. Differentially expressed genes (DEGs) were identified using Linear Models for Microarray Data (limma), followed by feature selection through three orthogonal ML models - Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Neural Network (NN). Overlapping markers among methods were used to define a consensus gene set, which was then compared with gene expression data from the Therapeutically Applicable Research to Generate Effective Treatments - Neuroblastoma (TARGET-NBL) cohort for clinical relevance assessment. We identified a seven-gene consensus set (BIRC5, CCDC127, AW473883, BRD7, TCF7L2, GABARAPL1, and CD1E) through ML, within which TCF7L2 emerged as a critical intersection between statistical and ML frameworks. While the large-scale 632-DEG set provided the broadest prognostic coverage, the refined seven-gene signature maintained robust clinical relevance. Notably, three mapped core genes (CD1E, GABARAPL1, and TCF7L2) were independently validated as significant favorable survival-associated markers (p < 0.05) in the TARGET cohort, even after rigorous clinical adjustment for age and COG risk status. By bridging transcriptomic discovery with cross-platform validation, this study establishes a high-confidence consensus gene signature for NB. Our findings suggest that while high-dimensional signatures offer extensive predictive power, the consensus-based core genes - particularly TCF7L2 - represent reproducible candidate biomarkers associated with disease progression and patient outcome.","[""Liu Daniel"", ""Kuo Jimmy"", ""Wu Yao-Chong"", ""Hsieh Shang-Yu"", ""Huang An Rong"", ""Lin Chorng-Horng""]",Computational biology and chemistry,2026-09-19,10.1016/j.compbiolchem.2026.109415,"[""Journal Article"", ""Meta-Analysis""]","[""Humans"", ""Neuroblastoma"", ""Machine Learning"", ""Oligonucleotide Array Sequence Analysis"", ""Gene Expression Profiling"", ""Computer Simulation"", ""Biomarkers, Tumor""]", 42800369,https://pubmed.ncbi.nlm.nih.gov/42800369/,Context-aware causal reasoning for explanatory visual question answering.,"Explanatory Visual Question Answering (EVQA) is a multimodal reasoning task that answers visually relevant natural language questions and generates user-friendly multimodal explanations. Although current EVQA methods can generate grammatically sound and vocabulary appropriate explanations, they still face the challenge of explanatory illusion (i.e., wrong explanations arrive at correct answers). To address this issue, we propose a novel Context-Aware Causal Reasoning (CACR) algorithm. Specifically, we first leverage a Multimodal Large Language Model and design prompting strategies to generate high-quality image descriptions as complementary evidence for helping generate correct explanations. Then, we propose a Structural Causal Model (SCM) to establish the causal relationship among the description, the explanation, and the answer. Finally, we transform the SCM into a deep variational inference network framework that enables causal reasoning to ensure that the explanations are grounded in the visual evidence related to the answer. Extensive experiments show that CACR significantly reduces the risk of explanatory illusion compared to six state-of-the-art methods, outperforming the best baseline by 2.22% on average in terms of explanation quality.","[""Miao Jiali"", ""Huang Xiaoling"", ""Yu Kui"", ""Shi Yecheng"", ""Wang Xiang"", ""Cao Fuyuan""]",Neural networks : the official journal of the International Neural Network Society,2026-09-23,10.1016/j.neunet.2026.109673,"[""Journal Article""]","[""Humans"", ""Algorithms"", ""Large Language Models"", ""Illusions"", ""Neural Networks, Computer""]", 42800368,https://pubmed.ncbi.nlm.nih.gov/42800368/,Foreground guided dual branch learning for sparse 4D radar 3D object detection.,"4D imaging radar offers long-range perception and Doppler velocity measurement capabilities, while maintaining strong robustness in nighttime scenes and adverse weather conditions such as rain and fog. These advantages make it a promising sensor for 3D perception in autonomous driving. However, due to radar imaging mechanisms, 4D radar point clouds are usually highly sparse and unevenly distributed, and are often contaminated by background clutter and isolated noise points. As a result, real objects are represented by only a few valid points and can be easily overwhelmed by complex background responses. Existing methods mainly focus on global feature enhancement or local denoising, while paying insufficient attention to explicit modeling of point-wise foreground characteristics. To address these issues, this paper proposes a foreground guided dual branch learning framework for 3D object detection in sparse 4D radar scenarios. Specifically, a differentiable learning-based query generator jointly models local density and motion consistency to estimate point-wise foreground importance and generate foreground aware cues. Then, a foreground guided gated cross branch fusion mechanism selectively injects auxiliary information into potential object regions, thereby enhancing object responses and suppressing background interference. Finally, at the proposal stage, cylindrical geometric constraints and a foreground prioritized keypoint selection strategy are combined to improve local feature representations. Experimental results on VoD and TJ4DRadSet demonstrate that the proposed radar-only method achieves competitive performance, even comparable to several radar-camera fusion methods, while maintaining a favorable balance between detection accuracy and inference efficiency.","[""Wang Yuanhang"", ""Zhou Yonghua"", ""Zhang Yongnan"", ""Mei Yiduo""]",Neural networks : the official journal of the International Neural Network Society,2026-09-23,10.1016/j.neunet.2026.109687,"[""Journal Article""]","[""Radar"", ""Detection Algorithms"", ""Imaging, Three-Dimensional"", ""Humans"", ""Algorithms"", ""Machine Learning""]", 42790277,https://pubmed.ncbi.nlm.nih.gov/42790277/,BFPSM: A biological language model-based framework for pathogenic synonymous mutation prediction.,"Accurate prediction of pathogenic synonymous mutations is essential for advancing precision medicine. To address this challenge, we propose BFPSM, a deep learning framework that systematically integrates heterogeneous biological language models for pathogenic synonymous mutations prediction based on sequences only. Specifically, BFPSM extracts embeddings from three DNA language models and one RNA language model to capture evolutionary conservation, long-range dependencies, reverse-complement information and splicing-related information, integrating them via feature concatenation to capture diverse biological contexts. Additionally, to fully exploit these multi-scale embeddings, high-dimensional outputs are processed by dynamic multi-scale convolutional neural network-bidirectional long short-term memory and lower‑dimensional ones by text convolutional neural network. This design enables multi-view feature extraction and enhances prediction robustness. Evaluated on the complete test set, BFPSM achieved an AUC of 0.928 and an AUPR of 0.941. On the largest common subset of the test set, the corresponding AUC and AUPR were 0.938 and 0.949, respectively. These results demonstrate that BFPSM outperforms state-of-the-art methods at the sequence level, highlighting its utility for clinical and functional variant prioritization. The source code has been made available on https://github.com/xialab-ahu/BFPSM.","[""Chen Hai"", ""Yang Siyuan"", ""Ye Chen"", ""Wang Chenlong"", ""Xia Junfeng"", ""Cheng Na""]",Computational biology and chemistry,2026-09-19,10.1016/j.compbiolchem.2026.109432,"[""Journal Article""]","[""Deep Learning"", ""Humans"", ""Silent Mutation"", ""Prediction Algorithms"", ""Mutation""]", 42790124,https://pubmed.ncbi.nlm.nih.gov/42790124/,Physics-informed adjoint neural network for parametric PDE-constrained optimal control problems.,"Many engineering applications involve solving parametric optimal control problems governed by partial differential equations (PDEs). Conventional numerical approaches suffer from high computational costs in multi-parameter scenarios, as they require solving the PDE system repeatedly for each parameter configuration, and remeshing may be also needed if domains vary with parameters. While deep learning offers a promising alternative by handling multiple parameters simultaneously, existing neural networks (NNs) still face challenges in accuracy and computational complexity. We propose two novel physics-informed adjoint NN (PIANN) methods for solving parametric optimal control problems. The PIANN1 employs a unified NN architecture that simultaneously approximates state and adjoint functions, optimized via optimality-condition loss minimization. The control solution is derived analytically from the adjoint-enhanced PDE system. The PIANN2 implements a coupled dual-network framework, where separate networks learn state and adjoint variables through an iterative direct-adjoint training process that inherently preserves constraints without penalty terms from Karush-Kuhn-Tucker (KKT) conditions. Both the PIANN1 and 2 simplify the network structure in comparison with the conventional NNs, and the PIANN2 avoids the involvement of penalty parameters for the KKT condition. The performance of these methods is verified by many benchmark parametric optimal control problems posed with the control constraint parameters, geometric parameters, corner singularities, and 3D domains. Numerical experiments show that the proposed methods achieve significantly higher accuracy compared to the existing NN techniques. In the appendix we further provide a theoretical analysis of the proposed methods, establishing residual-to-solution stability, convergence of the PIANN2 inexact direct-adjoint iteration under a small-gain condition, and a generalization bound that relates the empirical KKT loss to the continuous residual with explicit sample complexity. These results establish that the empirical residual minimization is a mathematically principled approach to solving parametric PDE-constrained optimal control problems.","[""Zeng Shaojie"", ""Han Zhaoxing"", ""Zhang Qinghui""]",Neural networks : the official journal of the International Neural Network Society,2026-09-21,10.1016/j.neunet.2026.109668,"[""Journal Article""]","[""Neural Networks, Computer"", ""Algorithms"", ""Soft Computing"", ""Deep Learning"", ""Computer Simulation""]", 42790123,https://pubmed.ncbi.nlm.nih.gov/42790123/,A broad learning model with dual-path feature encoding for robust and expedient incremental fault diagnosis.,"Intelligent fault diagnosis (IFD) models that rely on one-shot learning often struggle in dynamic mechanical systems because they exhibit limited knowledge stability and high computational cost under incremental learning (IL). To address the trade-off between robustness and efficiency, this paper proposes a Multi-Scale High-Resolution and Source-Informed Broad Learning Model (MHRSI-BLM). By leveraging the computational efficiency of broad learning for rapid weight correction and pseudo-inverse updating, the proposed framework achieves efficient incremental updates. In addition, the MHRSI-BLM framework adopts a dual-path feature-encoding strategy to improve robustness: a multi-scale high-resolution pathway enhances feature representation, while a complementary source pathway provides a relatively stable feature anchor for incremental updates. Experiments on two rolling-bearing datasets demonstrate that the proposed method achieves strong diagnostic accuracy with competitive computational efficiency.","[""Zhang Shengjie"", ""Xu Baoyi"", ""Yang Zeyun"", ""Wang Fen"", ""Liu Yuan"", ""Xiang Zheng""]",Neural networks : the official journal of the International Neural Network Society,2026-09-22,10.1016/j.neunet.2026.109677,"[""Journal Article""]","[""Machine Learning"", ""Algorithms"", ""Soft Computing"", ""Neural Networks, Computer"", ""Humans""]", 42785138,https://pubmed.ncbi.nlm.nih.gov/42785138/,"xLSTM-based deep learning model for predicting anticancer Activity of pyrano[3,2-a]phenazine hybrid derivatives: Proof of concept.","Finding novel treatments is a long and complex process involving extensive drug discovery, experimental testing, and preclinical research. Computational quantitative structure-activity relationship (QSAR) modeling can support this process by enabling rapid prediction of biological activity and identification of molecular features associated with activity. This study presents a proof-of-concept deep-learning-based QSAR framework for predicting the biological activity of pyrano[3,2-a]phenazine hybrid derivatives. The proposed framework combines an Extended Long Short-Term Memory (xLSTM) architecture with a hybrid APOADAM optimization strategy, in which Arctic Puffin Optimization (APO) is used for molecular descriptor subset selection and the Adam optimizer is used for xLSTM parameter optimization. Molecular descriptors were generated using the PaDEL-Descriptor tool and subjected to preprocessing, mutual-information-based ranking, and optimizer-based feature selection. The framework was evaluated for four cancer cell lines: HCT116 (colon), MCF7 (breast), HepG2 (liver), and A549 (lung). Model development was performed using 25 compounds with five-fold cross-validation, while 12 compounds were retained as a completely independent test set. On the independent test sets, the APOADAM-xLSTM model achieved RMSE values of 0.2514, 0.1712, 0.2451, and 0.3314 for HCT116, MCF7, HepG2, and A549, respectively. The corresponding Pearson correlation coefficients were 0.8252, 0.9577, 0.9314, and 0.8249. APOADAM also identified cell-line-dependent molecular descriptor subsets, indicating that different molecular features may contribute to biological activity across the investigated cell lines. Additional validation using out-of-fold predictions included Y-scrambling and applicability-domain analysis. The results provided evidence against chance correlations for three cell lines, while Williams-plot analysis showed that most OOF compounds were within the defined applicability domain. Overall, the results demonstrate the potential of the xLSTM-APOADAM framework for QSAR modeling under a constrained-data setting, while the findings remain preliminary and specific to the investigated chemical series. Further validation using larger and more chemically diverse datasets, broader benchmarking, and prospective experimental evaluation is warranted.","[""Aly Saad Aly Mohamed"", ""Maiyza Aya I"", ""Abuelmakarem Hala S"", ""Mohamed Sarah"", ""Fournier-Viger Philippe"", ""Khan Md Zaved H""]",Computational biology and chemistry,2026-09-22,10.1016/j.compbiolchem.2026.109431,"[""Journal Article""]","[""Humans"", ""Antineoplastic Agents"", ""Quantitative Structure-Activity Relationship"", ""Phenazines"", ""Deep Learning"", ""Cell Line, Tumor"", ""Drug Screening Assays, Antitumor"", ""Long Short Term Memory"", ""Proof of Concept Study"", ""Cell Proliferation"", ""Molecular Structure""]", 42777495,https://pubmed.ncbi.nlm.nih.gov/42777495/,Topological suppression of octahedral LiF networks enables fast Li+ transport in LiF-rich interphases.,"LiF-rich solid electrolyte interphases (SEIs) are widely associated with stable battery cycling, despite the intrinsically poor Li+ conductivity of crystalline LiF. Here, we investigate how crystallization and the temporal dynamics of local Li-F coordination states regulate ion transport in LiF-Li2CO3 inorganic interphase models. A fine-tuned neuroevolution potential enables 10 ns machine-learning molecular dynamics (MLMD) simulations of systems containing approximately 13,000 atoms. Cooling pure LiF from 1000 to 800 K induces rock-salt-like ordering, a pronounced increase in octahedral coordination, and a three-order-of-magnitude decrease in the long-time Li+ diffusivity. Motif-resolved survival and interconversion analyses reveal that this transport suppression is associated not only with the instantaneous octahedral population, but also with the emergence of long-lived octahedral states and an approximately 200-fold reduction in tetrahedral-octahedral exchange. Representative climbing-image nudged elastic band pathways yield migration barriers of 0.75 and 1.65 eV for tetrahedral-like and octahedral coordination environments, respectively, supporting the greater kinetic constraint imposed by octahedral coordination. Incorporation of Li2CO3 suppresses LiF crystallization, disrupts persistent octahedral networks, and restores dynamically exchangeable non-octahedral environments. The resulting diffusivity varies non-monotonically with carbonate content, while the highest-diffusivity composition shifts toward a higher carbonate content upon cooling. These results extend static coordination descriptions by identifying coordination-state persistence and exchange kinetics as dynamic atomistic descriptors of ion transport in heterogeneous LiF-rich inorganic interphases.","[""Fan Huiyang"", ""Yan Jianhua"", ""Bo Zheng"", ""Fan Zheyong"", ""Liu Zhu""]",Journal of colloid and interface science,2026-09-18,10.1016/j.jcis.2026.141620,"[""Journal Article""]",, 42767134,https://pubmed.ncbi.nlm.nih.gov/42767134/,"Network toxicology, transcriptomics, immune infiltration, and machine learning-based construction of an environmental endocrine disruptor exposure-associated predictive model for osteonecrosis of the femoral head.","To investigate the potential molecular links between nonylphenol (NP) exposure and osteonecrosis of the femoral head (ONFH), we employed an integrated strategy combining network toxicology, transcriptomic analysis, weighted gene co-expression network analysis (WGCNA), machine learning, immune infiltration analysis, molecular docking, molecular dynamics simulation, and independent dataset validation. Based on the GSE123568 discovery dataset of steroid-induced ONFH and steroid-exposed non-ONFH controls, 391 differentially expressed genes were identified, including 123 upregulated and 268 downregulated genes. Intersection of DEGs with WGCNA-derived module genes yielded 342 SONFH-associated candidate genes for subsequent analysis. NP-related targets were collected from CTD, SwissTargetPrediction, and Target Net databases, and 40 overlapping genes were identified as potential NP-related ONFH targets. PPI network analysis highlighted 10 key genes, while LASSO regression, random forest, and SVM-RFE identified PECAM1, CXCR2, and MAP2K3 as common core feature genes. Integrated candidate genes were mainly associated with inflammation, oxidative stress, vascular regulation, extracellular matrix remodeling, endocrine-lipid metabolic disturbance, and bone metabolism. ROC analysis showed that the candidate gene set discriminated SONFH samples from steroid-exposed non-ONFH controls in the discovery dataset, with an AUC of 0.967. Immune infiltration analysis revealed that PECAM1 was significantly correlated with multiple immune cell subsets. Molecular docking and a 100 ns molecular dynamics simulation provided exploratory structural support for a potential NP-PECAM1 interaction. In summary, PECAM1 may represent a prioritized candidate node linking NP-related molecular perturbations with SONFH/ONFH-associated endothelial-immune dysregulation.","[""Zhou Jiaxiang"", ""Yuan Huisheng"", ""Kuang Xiudong"", ""Wang Zhikai"", ""Chen Hongkun"", ""Du Qian"", ""Fan Qinghong"", ""Li Jianfeng"", ""Gao Manman"", ""Zhou Zhiyu""]",Computational biology and chemistry,2026-09-17,10.1016/j.compbiolchem.2026.109422,"[""Journal Article""]","[""Humans"", ""Endocrine Disruptors"", ""Machine Learning"", ""Molecular Docking Simulation"", ""Transcriptome"", ""Gene Expression Profiling"", ""Femur Head Necrosis"", ""Molecular Dynamics Simulation"", ""Phenols""]", 42766880,https://pubmed.ncbi.nlm.nih.gov/42766880/,A deep learning-assisted whole-cell biosensor for catechol monitoring: Synergizing BphC recognition with a multi-feature fusion vision transformer.,"Catechol (Cat) quantification typically requires benchtop instrumentation, restricting its application for on-site environmental analysis. To address this, we developed a portable image-based biosensing platform that integrates a genetically engineered whole-cell biocatalyst with a Multi-Feature Fusion Vision Transformer (MF-ViT). An Escherichia coli strain expressing 2,3-dihydroxybiphenyl-1,2-dioxygenase (BphC) was constructed to specifically convert Cat into a measurable yellow product. Molecular docking and molecular dynamics simulations provided a structural rationale for the sensor's analytical selectivity, demonstrating that specific bidentate coordination requirements prevent structural analogs from triggering false-positive signals. For signal decoding under ambient lighting, the MF-ViT model combined global visual representations with predefined colorimetric statistical features derived from the solution and background regions. This approach outperformed conventional simple RGB and convolutional neural networks (CNNs) analysis, achieving a high predictive accuracy (R2 = 0.944). The integrated biosensor exhibited a linear range of 12.5 to 400 nM and a limit of detection of 1.66 nM. Spiked recovery tests in seawater and tap water demonstrated that the smartphone-derived predictions were statistically consistent with standard spectrophotometric measurements. This approach provides a practical, algorithm-assisted strategy for the rapid screening of phenolic pollutants.","[""Qi Xiaoyan"", ""Sun Guoqiang"", ""Xue Yuzhi"", ""Liang Yan"", ""Wang Pingyu"", ""Ning Fangling"", ""Wu Dongyang"", ""Zhang Wenzhen"", ""Li Youjie"", ""Xie Shuyang""]",Talanta,2026-09-18,10.1016/j.talanta.2026.130638,"[""Journal Article""]","[""Biosensing Techniques"", ""Catechols"", ""Deep Learning"", ""Escherichia coli"", ""Molecular Docking Simulation"", ""Molecular Dynamics Simulation"", ""Dioxygenases"", ""Convolutional Neural Networks"", ""Colorimetry""]", 42762755,https://pubmed.ncbi.nlm.nih.gov/42762755/,Colorimetric sensor array based on pH-modulated inhibitory effect of a single nanozyme'activity for antioxidant recognition.,"To address the need for simultaneous identification of different antioxidants in complex samples, we constructed a simple and effective nanozyme-based colorimetric sensor array. The sensor array employs GaZn nanozymes with peroxidase-like activity as sensing units. Under three distinct pH conditions, the inhibitory effects of antioxidants on the enzymatic activity of the GaZn nanozyme were differentially modulated. Consequently, the nanozyme generates pH-dependent colorimetric responses, producing a unique signal pattern for each antioxidant. In this study, cysteine (Cys), glutathione (GSH), gallic acid (GA), chlorogenic acid (CA), and ascorbic acid (AA) were selected as target analytesto verify the recognition ability of our designed sensor array. Linear discriminant analysis (LDA) was applied to reduce the dimensionality of the colorimetric response data, achieving clear discrimination among the five substances, while a correlation between colorimetric response intensity and concentration was also observed. On this basis, further combined with machine learning algorithms such as random forest, the accuracy of the model in distinguishing blind samples and actual water samples has been effectively improved.","[""Feng Xinyi"", ""Liang Shuang"", ""Gu Dongxue"", ""Leng Yumin"", ""Chen Zhengbo""]",Talanta,2026-09-18,10.1016/j.talanta.2026.130637,"[""Journal Article""]","[""Colorimetry"", ""Antioxidants"", ""Hydrogen-Ion Concentration"", ""Glutathione"", ""Ascorbic Acid"", ""Machine Learning"", ""Gallic Acid"", ""Cysteine"", ""Chlorogenic Acid"", ""Discriminant Analysis"", ""Biosensing Techniques""]", 42759314,https://pubmed.ncbi.nlm.nih.gov/42759314/,Laser-induced graphene arrays for high-throughput metabolic profiling reveal serum metabolic alterations in colon cancer.,"Given the invasiveness of colonoscopy and the limited sensitivity and specificity of carcinoembryonic antigen for colon cancer detection, we developed a laser desorption/ionization mass spectrometry (LDI-MS) method based on a polyethersulfone film-based laser-induced graphene (PES-LIG) array. The PES-LIG is fabricated by one-step CO2 laser scribing on a PES film and can be directly used as an LDI substrate without any modification. Its layered structure facilitates the desorption/ionization of small-molecule metabolites, enabling uniform detection (RSD = 4.12%) with fmol-level sensitivity, high salt tolerance (S/N = 5319.8 for 1 nmol cellobiose in saturated NaCl solution), and low background. This method requires no matrix mixing, and a single circular LIG spot of 3 mm diameter can be produced in only ∼12 s. In a single experiment, a single PES film (76 mm × 25 mm) yielded a total of 94 LIG spots, with a material cost of only ∼0.04 USD. Thus, this approach enables rapid and high-throughput mass spectrometric analysis. Through computer-controlled laser scanning, PES-LIG arrays can be rapidly prepared and were applied to serum metabolomics of colon cancer. A total of 209 samples (106 colon cancer patients and 103 age-matched healthy controls) were analyzed, and 18 metabolites exhibited significant intergroup differences, revealing colon cancer-associated metabolic reprogramming. A neural network model constructed from these metabolic fingerprints achieved a diagnostic accuracy of 98.3% in blind validation. This work provides a low-cost, high-throughput strategy for non-invasive detection of colon cancer and offers a new high-performance substrate for LDI-MS technology.","[""Ye Zhihao"", ""Song Mengyang"", ""Ji Zhenyang"", ""Chen Junyu"", ""Nie Zongxiu"", ""Li Yuze""]",Talanta,2026-09-16,10.1016/j.talanta.2026.130614,"[""Journal Article""]","[""Humans"", ""Graphite"", ""Colonic Neoplasms"", ""Spectrometry, Mass, Matrix-Assisted Laser Desorption-Ionization"", ""Lasers"", ""Polymers"", ""Metabolomics"", ""Metabolome"", ""Sulfones""]", 42753521,https://pubmed.ncbi.nlm.nih.gov/42753521/,Bioinformatics-driven discovery and cross-cohort validation of a two-gene signature (RPH3A and HIGD1B) associated with Alzheimer's disease.,"BACKGROUND: Transcriptomic biomarker discovery for Alzheimer's disease (AD) has yielded numerous candidate signatures, yet many fail independent external validations due to overfitting or cohort-specific confounders. We aimed to identify a minimal reproducible gene signature and evaluate its generalizability under locked-model validation. METHODS: Three GEO datasets were analyzed: GSE118553 (training), GSE122063 (internal validation), and GSE5281 (external validation). Differentially expressed genes were identified in the training set, followed by LASSO and Random Forest feature selection, and intersecting genes were filtered for concordant fold-change direction and P < 0.05 across cohorts. An elastic net model was evaluated using locked training parameters by ROC, calibration, and decision curve analyses,with exploratory threshold optimization performed to assess model performance. RESULTS: Four genes overlapped between machine-learning methods, but only RPH3A and HIGD1B passed the consistency filter. The locked model achieved AUCs of 0.818 and 0.974 in the internal and external cohorts, respectively. Internal validation showed calibration drift (intercept -2.697, slope 0.705), resulting in 0.000 specificity at the locked threshold. Cohort-specific threshold optimization increased specificity to 1.000 with 0.607 sensitivity. External validation showed good calibration (intercept 1.065, slope 1.203) and favorable decision-curve performance. CONCLUSION: RPH3A and HIGD1B constitute a reproducible AD-associated molecular signature with robust discrimination across independent postmortem brain cohorts, regions, and platforms. As all datasets were postmortem brain tissue, these findings reflect disease-associated molecular alterations. Calibration, confusion matrix, and decision curve analyses further support model performance and translational relevance.","[""Chen Fuqiang""]",Computational biology and chemistry,2026-09-15,10.1016/j.compbiolchem.2026.109417,"[""Journal Article"", ""Validation Study""]","[""Alzheimer Disease"", ""Humans"", ""Computational Biology"", ""Cohort Studies"", ""Gene Expression Profiling""]", 42753519,https://pubmed.ncbi.nlm.nih.gov/42753519/,Leakage-safe machine learning evaluation of viral T-cell IFN-gamma response prediction using IEDB data.,"OBJECTIVE: To benchmark leakage-aware machine-learning methods for ranking viral T-cell peptides by qualitative interferon-gamma (IFN-gamma) assay outcome in public Immune Epitope Database (IEDB) records. METHODS: Recovered curated IEDB records were evaluated with exact peptide-disjoint, edit-distance < =2 component-disjoint, and publication-year temporal tests. Character n-gram models were tuned separately within each design, and uncertainty was estimated by grouped bootstrap. Context models excluded outcome-derived evidence variables and were tested using grouped ablations and three definitions of discordant peptide-context labels. Calibration and a frozen 8-million-parameter ESM-2 baseline were supporting analyses. RESULTS: The recovered curated dataset contained 31,502 assays from 17,336 peptides. Best sequence-only PR-AUC was 0.479 (95% CI 0.406-0.552) in the exact peptide-disjoint test, 0.404 (0.273-0.546) under component separation, and 0.151 (0.126-0.183) temporally. Leakage-screened context increased PR-AUC from 0.535 to 0.771 under exact separation and from 0.501 to 0.755 by component. After strict censoring of post-cutoff training evidence, temporal context PR-AUC increased from 0.324 to 0.363 (paired uplift 0.039; 95% CI 0.016-0.061); the direction persisted after excluding 662 tied outcomes and restricting to consistent labels. Frozen ESM-2 embeddings did not outperform the strongest classical model in any design. CONCLUSION: Sequence contains assay-associated ranking signal, but performance depends strongly on validation design and deteriorates temporally. Context improves retrospective prediction but may encode study structure and is not causal evidence. The framework supports benchmarking and candidate prioritization, not clinical or vaccine-efficacy prediction.","[""Montefusco-Pereira Carlos Victor""]",Computational biology and chemistry,2026-09-15,10.1016/j.compbiolchem.2026.109416,"[""Journal Article""]","[""Interferon-gamma"", ""Machine Learning"", ""T-Lymphocytes"", ""Humans"", ""Prediction Algorithms"", ""Predictive Learning Models"", ""Epitopes, T-Lymphocyte"", ""Immunoinformatics"", ""Peptides""]", 42753516,https://pubmed.ncbi.nlm.nih.gov/42753516/,ICM-MD: Integrating TM-specific features and MD-derived structures for accurate prediction of inter-chain contacts in α-helical transmembrane homodimers.,"Characterizing the interactions of alpha-helical transmembrane homodimers at the residue level is crucial for understanding their structure and function. However, most computational tools are designed for globular proteins and fail to translate to transmembrane (TM) proteins, largely due to the unique environment of the membrane and the limited availability of high-resolution structural data. To address this challenge, we present a data-centric machine learning framework that overcomes the scarcity of experimentally resolved TM homodimer structures, which are essential for developing a robust machine learning model. The proposed approach leverages MD-derived structural models as surrogate supervision for training the model. This model integrates sequence-based and structure-based features to enhance inter-chain residue contact prediction in TM homodimers. It also leverages a simple yet effective feed-forward neural network, designed to enhance model's interpretability and scalability. Comparative evaluation against state-of-the-art models, including DeepHomo1, DeepHomo2, Glinter, and DeepTMP, demonstrates that our method achieves improved performance. On a test set of eight alpha-helical TM homodimers, the model outperforms DeepHomo1 and DeepHomo2 with ΔPrecision@L = 42.2% and 43.9% respectively, surpasses Glinter by ΔPrecision@L = 34.6%, and achieves 7.5% higher precision compared to DeepTMP in the mean top L precision ranking metric.","[""Almalki Bander"", ""Liao Li""]",Computational biology and chemistry,2026-09-14,10.1016/j.compbiolchem.2026.109402,"[""Journal Article""]","[""Membrane Proteins"", ""Molecular Dynamics Simulation"", ""Machine Learning"", ""Protein Conformation, alpha-Helical"", ""Protein Multimerization""]", 42748891,https://pubmed.ncbi.nlm.nih.gov/42748891/,Application of ultrasonic technology in sheet joining: review article.,"Ultrasonic joining plays an important role in low-damage and high-efficiency manufacturing of sheets, owing to its advantages such as concentrated energy, small heat-affected zone, and no need for filler materials. This paper provides a summary from three aspects: ultrasonic joining processes, ultrasonic non-destructive testing (NDT), and the application of data-driven methods. In terms of ultrasonic joining, the main forms of ultrasonic application in sheet joining are reviewed according to the differences in ultrasonic mechanisms, among which the ultrasonic vibration effect has become an important performance enhancement means in various hybrid joining processes. With regard to ultrasonic NDT, in-situ, non-contact, and intelligent approaches represent the current major development directions. Concerning data-driven methods, this paper summarizes the current status of their applications in process optimization and process monitoring, and points out that weak cross-scenario transferability and insufficient generalization performance of existing models remain key bottlenecks. Finally, this paper outlines future development directions, including multi-energy field hybridization, machine learning-assisted process control, and high-power equipment development.","[""Liang Ye"", ""Wei Zhe"", ""Shi Ye"", ""Yan Ming"", ""Lei Lei""]",Ultrasonics,2026-09-12,10.1016/j.ultras.2026.108292,"[""Journal Article"", ""Review""]",, 42748632,https://pubmed.ncbi.nlm.nih.gov/42748632/,Sequence signatures for the molecular differentiation of pnictogen- binding bicyclic peptides.,"The molecular differentiation of chemically similar, thiophilic heavy pnictogens remains a key challenge in metallopeptide engineering. Here, an integrated framework combining k-nearest neighbors (KNN) classification, residue positional mapping, graph-theoretic co-occurrence networks, and sequence physicochemical profiling was applied to over 12,000 Cys-constrained bicyclic peptides to decode arsenic(III) and bismuth(III) selectivity. The best model achieved robust minority-class discrimination on unaugmented test data (Recall=0.85, AUPRC=0.75). Based on the statistically validated sequence signatures, As(III)-binding motifs favor extended conformations, higher hydrophobicity, and elevated partial specific volume. In contrast, Bi(III)-selective motifs favor flexible, hydrophilic loops enriched with aromatic and imidazole residues. These sequence signatures provide granular, actionable rules for the de novo design of selective biosensors, scavengers, and chelators for heavy element discrimination, as well as for fostering a deeper understanding on how proteins discriminate metals.","[""Janairo Jose Isagani B""]",Computational biology and chemistry,2026-09-15,10.1016/j.compbiolchem.2026.109412,"[""Journal Article""]","[""Bismuth"", ""Peptides, Cyclic"", ""Amino Acid Sequence""]", 42743799,https://pubmed.ncbi.nlm.nih.gov/42743799/,"MultiCardioFusionNet: A multimodal fusion model for predicting coronary heart disease integrating time-domain, frequency-domain, and clinical information.","Coronary heart disease (CHD) remains a leading cause of morbidity and mortality worldwide, highlighting the need for accurate, accessible, and cost-effective diagnostic approaches. This study proposes MultiCardioFusionNet, a multimodal deep learning framework that integrates electrocardiographic (ECG) signals and clinical information (gender, age, and body mass index) for intelligent CHD diagnosis. To capture complementary disease-related characteristics, a three-branch architecture is designed to jointly extract Frequency-Domain features, temporal features, and clinical representations from multimodal data. The resulting multimodal representations are integrated through an attention-guided fusion strategy to exploit complementary information across spectral, temporal, and clinical domains. The framework was developed and evaluated using real-world multicenter clinical data. In the internal validation cohort, MultiCardioFusionNet achieved an AUC of 0.9327, while independent external validation yielded an AUC of 0.9261, demonstrating promising generalizability capability. The proposed model consistently outperformed several representative deep learning methods, including Gated Recurrent Unit, Convolutional Neural Network, Convolutional Neural Network - Long Short-Term Memory, Convolutional Neural Network - Bidirectional Long Short-Term Memory and Convolutional Neural Network - Transformer (GRU, CNN, CNN-LSTM, CNN-BILSTM, and CNN-transformer). These results indicate that integrating electrophysiological and clinical information can substantially enhance CHD diagnosis. The consistent performance across the internal test cohort and the independent multi-center external validation cohort suggests that MultiCardioFusionNet may serve as a non-invasive decision-support approach for assisting CHD identification and risk assessment among clinically suspected patients undergoing further diagnostic evaluation.","[""Xu Tianbo"", ""Zhang Hechao"", ""Xu Hongzeng"", ""Xu Shanshan""]",Computational biology and chemistry,2026-09-09,10.1016/j.compbiolchem.2026.109398,"[""Journal Article""]","[""Humans"", ""Coronary Disease"", ""Deep Learning"", ""Electrocardiography"", ""Convolutional Neural Networks"", ""Female"", ""Male""]", 42743764,https://pubmed.ncbi.nlm.nih.gov/42743764/,Data augmentation by an electro-vibratory feedback improves signal interpretation by neural networks.,"Data-driven machine learning algorithms represent an alternative to numerical or analytical modeling of vibrating objects for the purpose of extracting their material or geometric properties. Their performance is conditioned by the quality of the training dataset, which should contain all possible production configurations of the object and measurement conditions. In this work, we propose a data-augmentation method based on manipulations of the vibration properties, rather than numerical signal post-processing. Instead of modifying the sample's material or geometry, we connect an additional sensor and actuator to the sample and use low-latency electronics to control the feedback signal. This delivers us an electro-mechanical system with tunable oscillation properties. This signal diversification method is demonstrated on the classification of correct and incorrect adhesion of two aluminum plates. The labeling accuracy of 67 % on nine unknown test objects, when trained on a single unchangeable training object, increases to 83 % when the variety of the training vibration signals is augmented by varying the parameters of the electro-vibratory feedback. This provides advantages for industrial quality control, where a large number of reference training samples are not available.","[""Rus Janez"", ""Tuuva Tim"", ""Fleury Romain""]",Ultrasonics,2026-09-08,10.1016/j.ultras.2026.108298,"[""Journal Article""]",, 42735448,https://pubmed.ncbi.nlm.nih.gov/42735448/,Parkinson's disease diagnosis using machine learning and multimodal biomarkers: A systematic review.,"Parkinson's disease (PD) is the second most common neurodegenerative disorder and the fastest-growing neurological condition globally, with an estimated 10.5 million people living with PD in 2026 and projections exceeding 14.2 million by 2040 according to the World Health Organization. Despite this staggering burden, diagnosis continues to depend on clinical motor signs that appear only after 60%-80% of dopaminergic neurons in the substantia nigra have been irreversibly lost, and specialist misdiagnosis rates reach 25% at early disease stages even in dedicated movement disorder centers. This systematic review synthesizes evidence from 239 studies drawn from 5847 screened records (2016-2026), critically evaluating unimodal diagnostic approaches encompassing dopamine transporter single-photon emission computed tomography (DAT-SPECT), quantitative magnetic resonance imaging (MRI), alpha-synuclein (α-syn) seed amplification assay (SAA), neurofilament light chain (NfL), genetic risk markers, voice and speech biomarkers, gait and movement metrics, electroencephalography (EEG), and handwriting analysis, alongside their integration into multimodal machine learning (ML) and deep learning (DL) frameworks with explainable artificial intelligence (XAI) components. Across 87 quantitative studies, the α-syn SAA achieved a sensitivity of 84%-92% and specificity of 93%-98% in prodromal cohorts. DAT-SPECT achieved a pooled sensitivity and specificity of approximately 90% but could not differentiate PD from atypical parkinsonian syndromes (APS). Multimodal fusion architectures, particularly cross-attention transformers integrating motor, voice, and gait data, consistently outperformed unimodal approaches by 4%-8% in the area under the receiver operating characteristic curve (AUC), with the largest gains in early stage detection and differential diagnosis tasks. XAI methods, including SHapley Additive exPlanations (SHAP), gradient-weighted class activation mapping (Grad-CAM), and integrated gradients, were reported in only 31% of deep learning studies. This indicates a transparency gap with direct regulatory consequences. Across all modalities, only 22% of studies reported external validation, and multicenter prospective evidence remains scarce. These two deficits, rather than model architecture, are now the principal barriers to clinical adoption. This review provides an actionable roadmap for researchers, clinicians, and regulatory agencies working in the PD diagnostics field.","[""Khan Muhammad Imran"", ""Imran Azhar"", ""Alshammari Ahmad""]",Computational biology and chemistry,2026-09-12,10.1016/j.compbiolchem.2026.109388,"[""Journal Article"", ""Systematic Review"", ""Review""]","[""Humans"", ""Parkinson Disease"", ""Machine Learning"", ""Biomarkers"", ""Dopaminergic Imaging""]", 42732722,https://pubmed.ncbi.nlm.nih.gov/42732722/,Data-driven discovery of polyoxometalate catalysts for selective ethanol conversion to ethyl acetate.,"The one-pot oxidation-esterification of ethanol to ethyl acetate (EA) provides a high-atom-economy route for converting bioethanol into a valuable product. Nonetheless, developing effective catalysts is challenging, as balancing EA selectivity with ethanol conversion is complex. Polyoxometalates (POMs) are promising multifaceted catalysts, but their vast chemical space makes identifying optimal catalysts through trial-and-error prohibitively difficult. Herein, we present a strategy that integrates machine learning (ML) and theoretical calculations to accelerate the discovery and understanding of POM catalysts. A dataset of 110 POMs is compiled, and an SVR-XGBoost stacking model is trained to predict POM activity. The model predicts ethanol conversions of up to 96.5% for KPW11Co and EA selectivity of 67.2% for KPW11Ni, which identifies them as superior candidates. Subsequent experiments validated these predictions. In-situ investigations and theoretical calculations reveal distinct reaction pathways for KPW11Co and KPW11Ni. The Co2+ center (3d7) facilitates the activation of O2via an oxidative pathway mediated by ·O2-. The Ni2+ center (3d8) primarily acts as a Lewis acid, coordinating with ethanol to promote direct dehydrogenation. KPW11Ni's distinct reactive pathway makes it exceptional for bioethanol conversion, achieving EA selectivity of up to 80.1%. This research develops a straightforward and reliable prediction framework that integrates descriptor-based ML screening with mechanistic verification, supporting the effective development of POM catalysts and the high-value conversion of alcohols.","[""Wang Jian"", ""Yan Jie"", ""Zhang Xiaoxuan""]",Journal of colloid and interface science,2026-09-09,10.1016/j.jcis.2026.141565,"[""Journal Article""]",, 42732656,https://pubmed.ncbi.nlm.nih.gov/42732656/,Synergistic selectivity carbon-dot-embedded imprinted fluorescence sensor with decision tree for intelligent on-site detection of hexavalent chromium.,"A novel synergistic selectivity carbon dot-embedded ion-imprinted fluorescence sensor was fabricated for intelligent on-site detection of hexavalent chromium (Cr(Ⅵ)). One-pot hydrothermal method synthesis yielded CDs that preferentially fluoresce toward Cr(VI) owing to surface amino, hydroxyl, and thiol groups. To further enhance the recognition selectivity, CDs were embedded into an ion-imprinted polymer via inverse microemulsion polymerization using 4-vinylpyridine as the functional monomer and ethylene glycol dimethacrylate as the cross-linker. The resulting carbon-dot-embedded ion-imprinted polymers (CDs@IIPs) combine the intrinsic selectivity of CDs with the specific recognition capability of the imprinted cavities, achieving a synergistic selectivity sensing platform. Mechanistic studies reveal that fluorescence quenching in the presence of Cr(Ⅵ) is primarily governed by an inner filter effect (IFE), while the imprinted sites ensure the exclusive recognition of Cr(Ⅵ) even in complex matrices. Under optimal conditions, the sensor exhibits a broad linear range of 0.01-50.0 mg/L and a low limit of detection of 0.012 mg/L. Its practical applicability is successfully validated in complex matrices (including seawater, fish, shrimp, and mussels), yielding robust recoveries of 95.3%-106.2%, with RSDs below 3.8%. Furthermore, a decision tree machine learning model was integrated to directly correlate the RGB image features with the Cr(Ⅵ) concentrations, enabling instrument-free, real-time visual quantification with a 95.5% accuracy. This study presents a cost-effective, intelligent, and field-deployable strategy for reliable trace Cr(Ⅵ) monitoring for environmental and food safety.","[""Li Qin"", ""Chen Yuge"", ""Chen Rui"", ""You Jinjie"", ""Liu Hua"", ""Shi Xizhi""]",Talanta,2026-09-11,10.1016/j.talanta.2026.130611,"[""Journal Article""]",, 42731440,https://pubmed.ncbi.nlm.nih.gov/42731440/,YOLOv10-PDTFN: An integrated ROI segmentation and transformer fusion framework for osteoporosis classification in knee radiographs.,"INTRODUCTION: Reduced bone mineral density and degeneration of bone microarchitecture are hallmarks of Osteoporosis, a degenerative skeletal condition that raises the risk of fractures. Due to the subjective nature of conventional radiography assessment and the potential for inter-observer variability, early diagnosis and osteoporosis grading remain clinical challenges. Although recent developments in deep learning have demonstrated promise in automating diagnostic tasks, it is still challenging to identify contextual structural patterns and minor trabecular alterations in radiographs. OBJECTIVE: This research suggests a unique deep learning framework for the multi-class classification of Osteoporosis from knee X-ray images in order to overcome these constraints. A Parallel-Dynamic Transformer Fusion Network (PDTFN) is proposed that combines advanced ROI detection with hierarchical spatial-sequence feature fusion for multi-class classification of normal, Osteopenia, and osteoporosis cases. METHODOLOGY: ROI localization is performed using an Improved YOLOv10 integrated with context-aware and region-adaptive modules to capture fine tibiofemoral details. The extracted joint regions are then analyzed by PDTFN, where a Parallel Spatial Transformer Unit (PSTU) models captures fine-grained and global structures, a Dynamic Sequential Convolutional Memory Unit (DSCMU) captures progressive bone density variations, and a Progressive Spatial-Sequential Attention Fusion (PSSAF) module refines discriminative features. Training is optimized using the AdaBelief optimizer, while Eigen-CAM visualizations highlight clinically relevant bone regions. The proposed PDTFN has better sensitivity and specificity than conventional CNN-based classifiers for differentiating between Osteopenia, Osteoporosis, and normal. Both fine trabecular textures and larger structural features are effectively captured by the model when dynamic sequential modeling and adaptive attention mechanisms are used. RESULTS: The model's emphasis on clinically significant tibiofemoral areas is validated by Eigen-CAM representations. The proposed model achieved 98.87% accuracy, 98.52% F1-score, 97.67% kappa, and 98.23% AUC. CONCLUSIONS: With increased accuracy, robustness, and interpretability in osteoporosis evaluation, the suggested approach shows great promise as a computer-aided diagnostic tool to assist radiologists in clinical practice.","[""Harikiran J"", ""Kishan S Ravi"", ""Channapragada Rama Seshagiri Rao"", ""Raju Yeligeti"", ""Kumar G Sai Chaitanya""]",Computational biology and chemistry,2026-09-09,10.1016/j.compbiolchem.2026.109399,"[""Journal Article""]","[""Humans"", ""Osteoporosis"", ""Radiography"", ""Deep Learning"", ""Knee""]", 42731265,https://pubmed.ncbi.nlm.nih.gov/42731265/,Parafilm-laminated microfluidic cloth devices: A durable graphdiyne-based nanozyme platform for machine learning-assisted antibiotic detection.,"Point-of-care testing (POCT) for antibiotic residues is in high demand for food safety supervision. However, the development of low-cost, durable, and easily fabricated analytical devices remains a challenge. Here, we reported a facile hot-pressing strategy to fabricate Parafilm-laminated microfluidic cloth-based analytical devices (P-μCADs) using a hydrophilic chemical fiber cloth substrate and commercial Parafilm as a hydrophobic barrier. Optimized at 105 °C and 6.8 MPa for 3 min, the molten Parafilm fully infiltrated the cloth fibers and formed well-defined and durable barriers with exceptional resistance to organic solvents and surfactants. The fabricated P-μCADs achieved a minimum functional hydrophilic channel width of approximately 420 μm and an ultra-low manufacturing cost (<0.05 RMB per device), demonstrating the potential for large-scale production. To distinguish the types and concentrations of antibiotics, a colorimetric detection array was constructed on the P-μCADs by integrating a peroxidase-mimicking graphdiyne-based nanozyme, coupled with linear discriminant analysis for data classification. Moreover, a support vector machine model was developed to analyze real milk samples. The model achieved high diagnostic performance, with accuracy, sensitivity, and specificity all reaching 95.0%. Featuring ultra-low cost, operational simplicity, and high chemical stability, the proposed P-μCADs platform holds great potential for POCT in resource-limited settings and offers a versatile strategy for rapid analytical screening.","[""Xiao Yuxin"", ""Zhang Xinqing"", ""Tang Xiaoshen"", ""Xian Siqi"", ""Kong Jijie"", ""He Huan""]",Talanta,2026-09-11,10.1016/j.talanta.2026.130608,"[""Journal Article""]","[""Anti-Bacterial Agents"", ""Animals"", ""Milk"", ""Machine Learning"", ""Lab-On-A-Chip Devices"", ""Colorimetry"", ""Microfluidic Analytical Techniques"", ""Graphite""]", 42731264,https://pubmed.ncbi.nlm.nih.gov/42731264/,Early diagnosis of cadmium stress in rice by intelligent profiling of multiple response indicators with portable Raman SERS and deep learning.,"Cadmium contamination severely affects rice growth, yield, and quality, making early stress monitoring essential for agricultural management and food safety. However, traditional detection methods are cumbersome and time-consuming, limiting their applicability to early stress diagnosis. This study developed a rapid and accurate approach for discriminating cadmium stress levels in rice. Arginine-modified flower-like silver nanoparticles (Ag NPs-Arg) were synthesized to enhance Raman signals associated with three stress-response indicators: salicylic acid (SA), malondialdehyde (MDA), and peroxidase (POD) activity. Quantitative prediction models for these physiological indicators and a stress-level discrimination model were established. Among the evaluated models, the CNN-Transformer model achieved the best predictive performance, with Rp2 values of 0.889, 0.832, and 0.802 for SA, MDA, and POD activity, respectively. An objective weighting method was used to integrate the three biochemical reference indicators, providing a multi-indicator physiological basis for comprehensive stress assessment. The resulting stress-level assessment model achieved an accuracy of 95.83%, demonstrating its ability to capture cadmium-induced physiological changes and assess stress levels in rice.","[""Wu Ting"", ""Zhang Leilei"", ""Ding Yanfei"", ""Sun Hongwei"", ""Gao Yutong"", ""Zhao Qianwen"", ""Tang Fan"", ""Lu Chenze"", ""Jiang Qiong"", ""Zhu Cheng""]",Talanta,2026-09-07,10.1016/j.talanta.2026.130586,"[""Journal Article""]","[""Oryza"", ""Cadmium"", ""Metal Nanoparticles"", ""Spectrum Analysis, Raman"", ""Silver"", ""Stress, Physiological"", ""Deep Learning"", ""Malondialdehyde"", ""Salicylic Acid"", ""Early Diagnosis"", ""Peroxidase""]", 42727360,https://pubmed.ncbi.nlm.nih.gov/42727360/,Machine learning-integrated molecular subtyping reveals two biologically distinct endometriosis subtypes in the EndometDB database.,"BACKGROUND: Endometriosis affects approximately 10% of reproductive-age women, with a diagnostic delay of 7-10 years. Despite clinical heterogeneity, current rASRM staging poorly predicts treatment outcomes. Molecular subtyping may reveal biologically meaningful patient strata that complement anatomical staging. OBJECTIVE: To identify molecular subtypes of endometriosis through integrated bioinformatics and construct machine learning (ML)-based diagnostic and prognostic models. METHODS: Gene expression data from EndometDB (GSE141549; 89 endometriosis patients, 41 controls) were processed. Differentially expressed genes (DEGs), weighted gene co-expression network analysis (WGCNA), and protein-protein interaction (PPI) networks were integrated to identify core genes. Consensus clustering defined molecular subtypes. ML diagnostic (XGBoost, ensemble) and prognostic (random survival forest) models were constructed and validated by nested cross-validation. An exploratory treatment response model was additionally evaluated. RESULTS: 823 DEGs (496 upregulated, 327 downregulated), 173 core genes, and 2 molecular subtypes were identified. Subtypes showed distinct ssGSEA pathway profiles but no statistically significant differences in rASRM score (p = 0.641) or age, suggesting molecularly-defined rather than clinically-defined heterogeneity. The ensemble diagnostic model achieved AUC= 0.892 (5-fold nested cross-validation). The random survival forest prognostic model, based on a surrogate endpoint, demonstrated an OOB concordance index of 0.751. An exploratory treatment response model performed no better than chance (AUC = 0.492). CONCLUSIONS: Integrating multi-method bioinformatics with ML identified 2 distinct endometriosis molecular subtypes and yielded internally validated diagnostic and prognostic models, providing an analytical framework and candidate molecular targets for future studies of endometriosis heterogeneity.","[""Wang Yiqun"", ""Cai Xiaozhen"", ""Xu Yi"", ""Ma Xia""]",Computational biology and chemistry,2026-08-26,10.1016/j.compbiolchem.2026.109349,"[""Journal Article""]","[""Endometriosis"", ""Humans"", ""Female"", ""Machine Learning"", ""Prognosis"", ""Random Forest"", ""Gene Expression Profiling"", ""Databases, Genetic"", ""Protein Interaction Maps"", ""Computational Biology"", ""Predictive Learning Models""]", 42727359,https://pubmed.ncbi.nlm.nih.gov/42727359/,Uncovering hub genes and key pathways responsive to drought stress in rice via meta-analysis of transcriptomic data.,"Drought stress presents a formidable threat to global rice cultivation, triggering complex molecular responses that impact plant growth and productivity. To decipher the underlying gene expression dynamics, we performed a comprehensive meta-analysis of transcriptomic datasets derived from drought-tolerant rice genotypes. Via microarray data from three independent studies, we identified a set of consistently expressed differentially expressed genes (DEGs) under drought conditions. Integration of functional annotation tools, including GO and KEGG pathway enrichment, revealed key biological processes and signaling cascades involved in stress mitigation, such as ABA signaling, protein folding, and photosynthesis suppression. Protein-protein interaction (PPI) network construction, followed by hub gene identification via maximal clique centrality (MCC), highlighted pivotal regulators including LEA proteins, dehydrins, HSP70, and several transcription factors. Machine learning approaches further prioritize potential biomarkers, with Random Forest models achieving high classification accuracy and pinpointing key predictive genes. Chromosomal localization analysis provided spatial insights into the distribution of these hub genes, whose expression patterns were further compared against qRT-PCR data from previously published studies. This integrative approach identifies candidate genomic markers and mechanistic insights that may support future breeding strategies for drought-tolerant rice, pending experimental validation.","[""Joshi Sachin"", ""Chaudhary Parneeta"", ""Mrinal Rishi"", ""Chauhan Ankita"", ""Pandey Niharika"", ""Gautam Sneh"", ""Lohani Pushpa""]",Computational biology and chemistry,2026-09-08,10.1016/j.compbiolchem.2026.109394,"[""Journal Article"", ""Meta-Analysis""]","[""Oryza"", ""Droughts"", ""Stress, Physiological"", ""Transcriptome"", ""Drought Resistance"", ""Gene Expression Profiling"", ""Gene Expression Regulation, Plant""]", 42721737,https://pubmed.ncbi.nlm.nih.gov/42721737/,"Integrated GC-MS, LC-MS/MS and FTIR-deep-learning analysis for chemical differentiation and rapid discrimination of four porcini mushroom species.","Porcini mushrooms have high culinary and commercial value, but species-dependent differences in their flavor-related chemical composition and rapid authentication remain insufficiently characterized. In this study, four wild porcini mushroom species, Boletus bainiugan, Butyriboletus roseoflavus, Lanmaoa asiatica, and Rugiboletus extremiorientalis, were investigated using an integrated analytical strategy combining GC-MS, LC-MS/MS, Fourier transform infrared spectroscopy (FTIR), chemometrics, deep learning, and SHAP-based model interpretation. In this workflow, GC-MS and LC-MS/MS were used as the metabolomic characterization layer to define species-associated chemical differences within the investigated sample set in volatile organic compounds (VOCs), organic acids, and amino acid-related metabolites, whereas FTIR was used as the rapid spectral acquisition layer for non-destructive classification. The metabolomic results identified 770 volatile organic compounds, 59 organic acids, and 67 amino acid metabolites, among which 200 VOCs, 27 organic acids, and 29 amino acid metabolites were screened as differential markers. These data revealed that the four porcini mushroom species differed markedly in aroma-active compounds and taste-related metabolites, including pyrazines, lipid oxidation-derived aldehydes/ketones, sulfur-containing volatiles, organic acids, and amino acid derivatives. FTIR spectra were then processed using different preprocessing methods and modeled using back propagation neural network (BPNN) and convolutional neural network (CNN) classifiers. Among the tested models, the standard normal variate (SNV)-CNN model achieved the best classification performance, with F1 scores of 0.96-1.00 for the four species. SHAP analysis further linked the key discriminative wavenumbers to characteristic molecular vibrations, mainly O-H, N-H, C-H, C-O, and C-C bands, which were consistent with the metabolomic differences revealed by GC-MS and LC-MS/MS. Therefore, this study establishes a two-level analytical platform in which chromatographic-mass spectrometric profiling provides chemical annotation, compositional interpretation, and hypothesis generation, while FTIR combined with deep learning enables rapid species discrimination. The proposed strategy offers both chemical insight and practical potential for quality evaluation of porcini mushrooms.","[""Zheng Chuanmao"", ""Li Jieqing"", ""Liu Honggao"", ""Wang Yuanzhong""]",Talanta,2026-09-04,10.1016/j.talanta.2026.130595,"[""Journal Article""]","[""Spectroscopy, Fourier Transform Infrared"", ""Gas Chromatography-Mass Spectrometry"", ""Volatile Organic Compounds"", ""Tandem Mass Spectrometry"", ""Deep Learning"", ""Liquid Chromatography-Mass Spectrometry"", ""Agaricales"", ""Metabolomics""]", 42721705,https://pubmed.ncbi.nlm.nih.gov/42721705/,Machine learning driven QSPR analysis of anti-blood cancer drugs using graph invariants.,"The quantitative description of drug molecules through their molecular structures is central to computational studies of drugs. This paper presents a machine learning-driven quantitative structure-property relationship (QSPR) model linking graph-based structural descriptors to the physicochemical properties of anti-blood cancer drugs. A dataset of 13, anti-blood cancer drug molecules was characterized using six degree-based topological indices: the first Zagreb, second Zagreb, Sombor, reduced Sombor, harmonic, and Randić indices. These graph-derived descriptors were used as molecular features to train and compare six machine learning algorithms, namely linear regression, ridge regression, lasso regression, AdaBoost, XGBoost, and gradient boosting, for predicting nine physicochemical properties. Model performance was assessed using MSE,MAE,RMSE, and the coefficient of determination R2, together with a graphical comparison of observed and predicted values. The novelty of this work lies in the systematic, comparative evaluation of multiple graph-based descriptors across conventional, regularized, and nonlinear ensemble learning paradigms, rather than reliance on a single descriptor or algorithm, thereby revealing how effectively structural information embedded in topological indices can be transferred into property prediction. Gradient boosting emerged as the most efficient predictor, achieving R2=0.9996 to 0.9999,MSE=0.0027 to 33.8688,MAE=0.0425 to 3.9874, and RMSE=0.0519 to 5.8197, across all nine properties, confirming the strength of nonlinear ensemble learning for this task. These results demonstrate that combining degree-based topological descriptors with machine learning provides an effective and generalizable computational framework for QSPR modeling of anti-blood cancer drugs.","[""Naz Kiran"", ""Bilal Hafiz Muhammad"", ""Ashraf Faraha"", ""Siddiqui Muhammad Kamran""]",Computational biology and chemistry,2026-09-07,10.1016/j.compbiolchem.2026.109374,"[""Journal Article""]","[""Machine Learning"", ""Quantitative Structure-Activity Relationship"", ""Antineoplastic Agents"", ""Humans"", ""Boosting Machine Learning Algorithms"", ""Prediction Algorithms"", ""Predictive Learning Models""]", 42711831,https://pubmed.ncbi.nlm.nih.gov/42711831/,A Dual Soft Attention Network for Automatic Sleep Staging via Single-Lead Electrocardiogram Signals.,"Polysomnography (PSG) is the current gold standard for sleep monitoring; however, it is complex, burdensome, and costly. In contrast, commercial health monitoring devices, such as the Apple Watch, offer sleep scoring at a low cost over a few days. This paper proposes a dual-attention mechanism-based sleep staging method using single-lead electrocardiogram (ECG) signals, named SleepECGNet. The method extracts multi-scale features by employing convolutional kernels of different sizes, thereby capturing subtle details in the ECG signals. The model then integrates an SE module and a Luong attention-based encoder-decoder architecture to filter redundant information and dynamically focus on different sleep stages. Experiments were conducted using the HMC, MGH, and SHHS1 datasets for sleep staging tasks and the Apnea-ECG corpus was employed to execute transfer learning strategies directed at sleep apnea recognition. Comprehensive ablation experiments were conducted to verify the effectiveness of each component. Moreover, an analysis of sequence duration demonstrated that expanding the input timeframe directly improves classification outcomes.","[""Qi Xuanhao"", ""Li Qinying"", ""Liu Hongze"", ""Cheng Yinlin"", ""Ling Bingo Wing-Kuen"", ""Hu Zhensheng""]",International journal of neural systems,2026-09-09,10.1142/S0129065727500110,"[""Journal Article""]","[""Humans"", ""Electrocardiography"", ""Sleep Stages"", ""Signal Processing, Computer-Assisted"", ""Polysomnography"", ""Deep Learning"", ""Sleep Apnea Syndromes"", ""Attention""]", 42710147,https://pubmed.ncbi.nlm.nih.gov/42710147/,A source-aware chemometric benchmark reveals transportability gaps in heterogeneous lipid nanoparticle data.,"Data-driven lipid nanoparticle (LNP) design increasingly relies on heterogeneous datasets assembled across experimental programs, but random-row validation may overstate performance for unseen sources. We developed a source-aware benchmark using 18,326 functional records, 50 coherent assay strata, and 39 publication-defined source domains. LightGBM, logistic regression, and ridge models were evaluated under random-row, exact-ionizable-lipid, descriptor-cluster, and complete source-domain holdout, with preprocessing confined to training folds and metrics aggregated by source. LightGBM source-macro ROC AUC decreased from 0.764 (95% CI, 0.732-0.796) under random-row validation to 0.585 (0.542-0.628) under source holdout, a paired mean loss of 0.179 (P = 2.5 × 10⁻¹¹). Continuous-rank correlation declined from 0.463 to 0.160, while top-10% candidate enrichment decreased from 2.61-fold to 1.39-fold. Exact-lipid and descriptor-cluster holdouts produced intermediate performance, indicating that chemical novelty did not reproduce complete source novelty. Results were robust to weighting, fold assignment, cluster count, and leave-one-source-out validation. Random-row performance alone therefore did not establish cross-source transportability. Fixed source-aware splits, source-level estimands, and machine-readable outputs provide a reusable framework for more reliable LNP prediction.","[""Wu Wenhan"", ""Zhang Jun"", ""Huang Zhen""]",Computational biology and chemistry,2026-09-04,10.1016/j.compbiolchem.2026.109360,"[""Journal Article""]",, 42710145,https://pubmed.ncbi.nlm.nih.gov/42710145/,Construction of efficacy prediction model for T-DXd treatment of HER2-positive and HER2-low expression breast cancer based on hematological indicators: A real-world study.,"BACKGROUND: This study explores hematological changes during trastuzumab deruxtecan (T-DXd) treatment in HER2-positive and HER2-low breast cancer patients, aiming to develop an efficacy prediction model for personalized clinical decisions. MATERIALS: A retrospective analysis of 114 patients treated with T-DXd at Beijing Chaoyang District Sanhuan Cancer Hospital (March 2023-December 2024) was conducted. Data included demographics, treatment history, and laboratory results. Efficacy was assessed via RECIST 1.1, with analyses using the Wilcoxon test and Kaplan-Meier method. RESULTS: HER2-positive patients had a progression-free survival (PFS) of 12.2 months vs. 10.2 months for HER2-low. Common adverse events included Nausea, primarily grade 1 or 2. T-DXd caused significant reductions in red blood cell count, hemoglobin, hematocrit, creatine kinase, and homocysteine (p < 0.05), along with increased red cell distribution width. Responders (CR+PR+SD) and progressors (PD) differed in platelets, plateletcrit, and homocysteine (p < 0.05). Stratification by estrogen receptor (ER) and progesterone receptor (PR) expression and Ki67 revealed distinct PFS outcomes. We aimed to predict the efficacy of T-DXd using pre-treatment hematological parameters and various machine learning models, including Random Forest, SVM, XGBoost, and LightGBM. The performance of the optimal models, as evaluated by leave-one-out cross-validation, yielded AUC values of 0.748 (XGBoost), 0.881 (SVM), and 0.830 (SVM) for models based on Complete Blood Count (CBC) parameters, biochemical markers, and their combination, respectively. CONCLUSION: Hematological toxicity is a recognized concern with T-DXd therapy. Our predictive model demonstrates strong performance in identifying patients suitable for treatment, highlighting that rigorous monitoring of hematological indicators is critical for mitigating risk and upholding safety standards in clinical practice.","[""Su Yanfang"", ""Sang Die"", ""Zhang Guoyong"", ""Liu Haiyan"", ""Zhang Yurong"", ""Fan Shanmin"", ""Zhang Jintao"", ""Li Shiyu"", ""Zhang Yanyan"", ""Feng Huachao"", ""Zhao Meiqing"", ""Zhao Longmei"", ""Yuan Peng"", ""He Binsheng"", ""Li Man""]",Computational biology and chemistry,2026-09-04,10.1016/j.compbiolchem.2026.109391,"[""Journal Article""]","[""Humans"", ""Female"", ""Breast Neoplasms"", ""Erb-b2 Receptor Tyrosine Kinases"", ""Trastuzumab"", ""Retrospective Studies"", ""Middle Aged"", ""Predictive Learning Models"", ""Adult"", ""Random Forest"", ""Aged"", ""Boosting Machine Learning Algorithms""]", 42705098,https://pubmed.ncbi.nlm.nih.gov/42705098/,CRISPGen: A deep generative framework for multi-objective CRISPR/Cas9 guide RNA design via Conditional Latent Diffusion and Dual-Critic Reinforcement Learning.,"MOTIVATION: The CRISPR-Cas9 system offers transformative potential for precision genome editing, yet its clinical translation remains constrained by the risk of unintended off-target double-strand breaks. While current discriminative models excel at evaluating pre-specified candidate guides, resolving the fundamental antagonism between on-target cleavage efficiency and off-target specificity within a fixed sequence search space remains a major challenge. RESULTS: We present CRISPGen, a unified deep generative framework that reframes sgRNA design as a multi-objective constrained sequence synthesis problem. It integrates (i) DNABERT-2 genomic-language embeddings, (ii) a conditional latent diffusion generator conditioned on a user-specified on-target efficiency target, and (iii) a dual-critic reinforcement-learning (RL) stage that couples a frozen on-target efficiency critic with a cross-attention off-target discriminator (validation Pearson R=0.8157) trained on a unified corpus of experimental off-target events from six detection platforms. Across 1000 generated sgRNAs, CRISPGen reduces the mean off-target discriminator score by 99.7% relative to the pre-RL baseline and, under an exhaustive whole-genome screen of all 302,631,056 NGG PAM sites in GRCh38, yields zero perfect-match and only 55 one-mismatch genomic hits. We further show, transparently, that the internal on-target critic saturates under RL optimization - an instance of Goodhart's Law - and therefore assess on-target viability using an independent external CRISPRon screen (mean 47.10/100). Repeating the RL fine-tuning stage under three random seeds (with the diffusion generator, DNABERT-2 embeddings, and off-target discriminator held fixed) yields a stable operating point across seeds. Full diversity, per-mismatch, and reproducibility statistics are reported in the Results. AVAILABILITY: Source code is available at https://github.com/malekpouri/CRISPGen; the pre-trained checkpoints and the 3,000,000-sequence library are hosted on Hugging Face (https://huggingface.co/malekpouri/CRISPGen-Checkpoints) and archived on Zenodo under DOI 10.5281/zenodo.21428641.","[""Malekpouri Mohammad"", ""Lotfi Somayeh""]",Computational biology and chemistry,2026-08-29,10.1016/j.compbiolchem.2026.109328,"[""Journal Article""]","[""CRISPR-Cas Systems"", ""Reinforcement Machine Learning"", ""RNA, Guide, CRISPR-Cas Systems"", ""Generative Artificial Intelligence"", ""Gene Editing""]", 42705097,https://pubmed.ncbi.nlm.nih.gov/42705097/,Generalizable multi-modal medical image segmentation model using Multi-head Gated Cross Attention fusion Encoder-based adaptive Trans-Mobile-Unet++ with consistency loss function.,"Automated medical image segmentation is an essential component of modern clinical diagnosis, enabling the precise delineation of anatomical features and abnormal tissue from various medical imaging techniques, including Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) scans. However, achieving reliable segmentation across multi-modal medical images remains difficult because of variations in imaging properties, feature distributions, and modality-specific information. Existing segmentation methods often exhibit limited generalization and insufficient cross-modal feature integration, resulting in reducing segmentation accuracy. To resolve these concerns, the proposed framework employs an advanced attention mechanism-driven deep learning network to effectively learn complex heterogeneous features in medical images. The process begins by collecting the required multi-modal images from standard public databases, which are then directly provided to the segmentation model for performing multi-modal segmentation. The segmentation process is executed using Multi-head Gated Cross Attention Fusion Encoder-based Adaptive Transformer Mobile-UNet++ with Consistency Loss Function (MGAFE-ATMU++-CLF) for accurate segmentation of multi-modal medical images. The proposed model incorporates a multi-head gated cross-attention fusion mechanism throughout the encoder to efficiently extract complementary information from different imaging modalities while preserving discriminative features. Furthermore, a Renovated Puma Optimizer (RPO) is designed to fine-tune the critical network hyperparameters for the MGAFE-ATMU++ framework, thereby enhancing the segmentation capability of the developed approach. The evaluation results indicate the introduced MGAFE-ATMU++-CLF framework delivers improved segmentation accuracy by effectively extracting heterogeneous features, integrating cross-modal information, and accurately delineating anatomical boundaries, highlighting its potential for reliable analysis of multi-modal medical images.","[""Neeraja Chinnamgari"", ""Reddy G Umamaheswara"", ""Thumbur Gowri""]",Computational biology and chemistry,2026-08-29,10.1016/j.compbiolchem.2026.109366,"[""Journal Article""]","[""Magnetic Resonance Imaging"", ""Humans"", ""Tomography, X-Ray Computed"", ""Image Processing, Computer-Assisted"", ""Deep Learning"", ""Algorithms""]", 42702094,https://pubmed.ncbi.nlm.nih.gov/42702094/,High-throughput pharmaceutical classification via short-range interferometric hyperspectral imaging.,"Short-wave infrared hyperspectral imaging (SWIR-HSI) enables rapid, non-destructive material characterization with high potential for process analytical technology and pharmaceutical screening. Here, we demonstrate high-throughput, full-field SWIR-HSI using Fourier transform spectroscopy (FTS) coupled with deep learning. Conventional FTS typically requires thousands of sampling points over a wide optical path difference (OPD) range to resolve spectral features. By directly processing raw interferograms with deep learning, bypassing conventional Fourier transformation, we achieved 98.9% pixel-level and 100% whole-tablet classification accuracy across 11 over-the-counter pharmaceutical products using only 200 data points acquired over a 10 μm mirror displacement (20 μm OPD). Under challenging independent-lot, multi-day testing conditions, the framework maintained 96.7% pixel-level and 100% whole-tablet accuracy. Latent-space visualizations confirmed that short-range interferograms retain sufficient chemical and physical information for rapid, robust pharmaceutical classification. These findings indicate that deep-learned short-range interferometry offers a promising, vibration-resilient approach for high-throughput, non-destructive quality control of solid dosage pharmaceuticals.","[""Nyakuchena Melisa"", ""Hasan Khaled"", ""Sung Yongjin""]",Talanta,2026-09-02,10.1016/j.talanta.2026.130550,"[""Journal Article""]","[""Interferometry"", ""Spectroscopy, Fourier Transform Infrared"", ""Hyperspectral Imaging"", ""Deep Learning"", ""Tablets"", ""Pharmaceutical Preparations"", ""High-Throughput Screening Assays"", ""Fourier Analysis""]", 42696953,https://pubmed.ncbi.nlm.nih.gov/42696953/,MOF-regulated aggregation-induced electrochemiluminescence for machine learning-assisted aptasensing of cancer marker.,"Electrochemiluminescence (ECL) biosensing requires luminophores with high emission efficiency and favorable interfacial reaction kinetics. Herein, a nanoconfined aggregation-induced emission luminophore, TH-UiO-66-NH2, is constructed by coordinating tetracarboxyl-functionalized tetraphenylethylene (TCTPE) within UiO-66-NH2. The resulting TH-UiO-66-NH2 shows an 8.3-fold enhancement in ECL intensity compared with pristine TCTPE. Mechanistic studies reveal that the rigid UiO-66-NH2 not only enhances the interfacial enrichment and electrochemical activation of K2S2O8, but also regulates the aggregation state of TCTPE within a spatially confined coordination environment, thereby suppressing nonradiative relaxation of the AIE luminophore. Taking advantage of the excellent ECL performance of TH-UiO-66-NH2, a signal ""on-off-on"" biosensing platform is further developed for carcinoembryonic antigen (CEA) detection. The proposed biosensor shows a linear relationship ranging from 10-13 to 10-7 g mL-1 with detection limit down to 2.138 × 10-14 g mL-1. Moreover, a random forest-assisted machine-learning model further improves the detection accuracy, demonstrating the potential of this intelligent ECL system for highly sensitive CEA detection. This work offers a reliable and intelligent analytical strategy for the precision screening and auxiliary diagnosis of cancer markers.","[""Liu Jinxin"", ""Wu Rui"", ""Cheng Hao"", ""Yang Zhen"", ""Wang Jun"", ""Xi Cheng"", ""Hu Xinyu"", ""Zhang Weiwei"", ""Xu Ning"", ""Li Chuanping""]",Talanta,2026-09-03,10.1016/j.talanta.2026.130572,"[""Journal Article""]","[""Biosensing Techniques"", ""Luminescent Measurements"", ""Carcinoembryonic Antigen"", ""Electrochemical Techniques"", ""Machine Learning"", ""Biomarkers, Tumor"", ""Humans"", ""Limit of Detection"", ""Aptamers, Nucleotide"", ""Stilbenes""]", 42691707,https://pubmed.ncbi.nlm.nih.gov/42691707/,Recent advances of colorimetric sensors based on laccase-mimicking nanozymes.,"Natural laccase exhibits poor stability, high cost, strict storage requirements and weak recyclability, which greatly limits its utility in colorimetric sensing and catalytic remediation. Laccase-mimicking nanozymes circumvent these shortcomings. Copper derivatives replicate laccase Cu(I)/Cu(II) redox via tailored coordination, whereas Ag-, Mn-, Ce-, Co-, Pt- and Fe-based nanozymes realize comparable phenolic oxidation using multivalent metal redox pairs and oxygen vacancies modulated by heterostructures and ligands. This review outlines recent advances in nanozyme-based colorimetric sensors. We categorize laccase-mimicking nanozymes into predominant Cu-based variants (amino acid-, peptide-, nucleotide-, MOF-, copper oxide- and bimetal-doped Cu hybrids) and complementary Ag-, Mn-, Ce-, Co-, Pt- and Fe-based analogues, and detail their synthesis, catalytic pathways and structural optimization. Two categories of colorimetric sensing mechanisms are summarized: catalytic signal mechanisms (direct catalytic activation, inhibition and activity-enhancement-based sensing) and target-recognition mechanisms (aptamer-mediated and molecular imprinting-based recognition). We review their reported applications in environmental monitoring, food safety screening and biomedical biomarker quantification targeting phenols, pesticides, heavy metals, mycotoxins, antibiotics and catecholamines. Notably, most existing work remains laboratory-based proof-of-concept research, lacking validated field and clinical practical implementations. We also identify key limitations, including inadequate matrix stability, narrow substrate specificity, intricate synthesis and low single-target throughput. Prospective strategies are proposed, covering biomimetic active-site engineering, machine learning-aided sensor arrays, multimodal signal transduction, scalable low-cost fabrication and smartphone-coupled portable detectors.","[""Chi Zhongmei"", ""Ma Jiahui"", ""Chu Shiqi"", ""Yan Lihe"", ""Liu Guocheng""]",Talanta,2026-09-01,10.1016/j.talanta.2026.130541,"[""Journal Article"", ""Review""]","[""Laccase"", ""Colorimetry"", ""Biosensing Techniques"", ""Biomimetic Materials"", ""Catalysis"", ""Copper"", ""Humans"", ""Nanostructures"", ""Oxidation-Reduction"", ""Environmental Monitoring""]", 42691622,https://pubmed.ncbi.nlm.nih.gov/42691622/,Beyond random splits: A hierarchical benchmark of transferability and reliability in PROTAC activity prediction.,"Computational prediction of PROTAC degradation activity (DC50) has attracted growing interest, yet the reliability of reported model performance remains poorly understood because sufficiently stringent evaluation protocols are rarely applied. Here, we present a hierarchical benchmark designed to expose evaluation pitfalls and quantify the transferability and reliability limits of current PROTAC predictors. Using a curated dataset of 2405 DC50 measurements spanning 22 target proteins and two E3 ligases (CRBN and VHL), we benchmarked classical machine learning (Random Forest, ExtraTrees, Ridge, PLS), gradient-boosted trees (XGBoost), nearest-neighbor retrieval baselines, protein negative controls, and a representative multi-modal deep learning ensemble (HybridMoECrossAttn) across Random, Scaffold, Leave-One-Target-Out (LOTO), and Leave-One-Family-Out (LOFO) splits. Under Random evaluation, a simple Random Forest + ECFP4 baseline achieved pooled R² = 0.693 ± 0.025, indicating that conventional models already approach the apparent ceiling under interpolation-oriented settings. However, all methods collapsed under LOTO (best R² = -0.012), revealing that much of the apparent progress in the literature reflects chemical-neighbor memorization rather than robust target-level generalization. We further show that target-wise error is significantly associated with continuous protein semantic proximity in ProtBERT space (Spearman ρ = -0.461, p = 0.047), whereas coarse family-level descriptors are uninformative. A four-quadrant failure taxonomy reveals that protein shift is more damaging than chemical novelty (MAE 1.09-1.12 vs. 0.85-0.97), and conformal prediction becomes severely overconfident under target extrapolation, with empirical 90% coverage dropping to 63.9-66.7%. These results reposition PROTAC prediction as a problem of transferability and reliability rather than leaderboard optimization and provide practical guidelines for future benchmark design.","[""Zhu Renguang"", ""Guo Guanghao"", ""Li Lulu"", ""Luo Sihan""]",Computational biology and chemistry,2026-08-27,10.1016/j.compbiolchem.2026.109358,"[""Journal Article""]","[""Proteolysis Targeting Chimera"", ""Benchmarking"", ""Proteins"", ""Reproducibility of Results"", ""Prediction Algorithms"", ""Machine Learning"", ""Humans""]", 42691620,https://pubmed.ncbi.nlm.nih.gov/42691620/,From MMP cliffs to binding interactions: An integrated Read-across and deep learning-based investigation of MMP-12 inhibitors to elucidate S1' pocket recognition.,"Matrix metalloproteinase-12 (MMP-12) is a zinc-dependent endopeptidase that plays an important role in the pathogenesis of several inflammatory, pulmonary, cardiovascular, neurological, and cancer-associated disorders. Despite its therapeutic significance, the development of potent and selective MMP-12 inhibitors remains challenging because of the high structural similarity shared among various MMP family members. This study presents an integrated framework combining matched molecular pair (MMP) cliff analysis, quantitative read-across structure-activity relationship (qRASAR)- based modelling, deep learning-based analysis of binding interactions, and MD simulations to elucidate the structural determinants governing potent MMP-12 inhibition, with particular emphasis on recognition of the S1' pocket. Leveraging structural similarity with MMP-12 inhibitors, the final qRASAR MLR model showed satisfactory predictive performance (R2 = 0.710, Q2F1 = 0.734, Q2F2 = 0.734, and MAEtest = 0.563). A physics-aware, deep learning-based binding interaction analysis showed that potent inhibitors like C5 and C54 form favourable interactions with key residues in the S1' pocket, including P238, Y240, K241, and F248, whereas weak inhibitors like C475 exhibit comparatively weaker engagement within this S1' subsite. Subsequently, MD simulations further confirmed the enhanced stability, compactness, and reduced conformational flexibility of the MMP-12-C5 and MMP-12-C54 complexes relative to MMP-12-C475 and highlighted the critical role of persistent S1' pocket interactions in stabilizing the protein-ligand complexes and enhancing MMP-12 inhibitory potency. The findings emphasize the importance of effective S1' pocket recognition and provide valuable insights for the rational design of potent MMP-12 inhibitors in future.","[""Dasgupta Indrasis"", ""Sensarma Asmita"", ""Amin Sk Abdul"", ""Concilio Simona"", ""Basak Piyali"", ""Piotto Stefano"", ""Gayen Shovanlal""]",Computational biology and chemistry,2026-08-31,10.1016/j.compbiolchem.2026.109375,"[""Journal Article""]","[""Matrix Metalloproteinase 12"", ""Deep Learning"", ""Matrix Metalloproteinase Inhibitors"", ""Molecular Dynamics Simulation"", ""Humans"", ""Binding Sites"", ""Quantitative Structure-Activity Relationship"", ""Protein Binding""]", 42691619,https://pubmed.ncbi.nlm.nih.gov/42691619/,A large-scale cryo-EM RNA motif dataset and benchmark for machine learning-based structure modeling.,"MOTIVATION: RNA functions in gene regulation, viral replication, and cellular control are tightly coupled to three-dimensional structure and local conformational features. Cryogenic electron microscopy (cryo-EM) now enables RNA structure characterization across a broad resolution range, but full maps are large, heterogeneous, and variable in local resolution. RNA secondary structural motifs, including hairpins, internal loops, and bulges, provide recurring local units for interpreting RNA density, comparing structures, and developing machine-learning models. Existing cryo-EM-based methods generally focus on complete maps, chains, residues, or atomic model construction rather than motif-level representations, partly because large-scale motif-resolved cryo-EM datasets remain limited. RESULTS: We present an open-source dataset of more than 100,000 motif-resolved cryo-EM density segments paired with atomic structures, spanning 25 RNA secondary structural motif classes and resolutions from 1.5 Å to 34.0 Å. Each motif is represented as a standardized 3D voxel grid with voxel-level labels for RNA backbone, ribose sugar, and nucleobase components. Motif-level map-model agreement was evaluated using masked cross-correlation (CCmask) and atom-level Q-scores, revealing resolution-dependent trends in regional density agreement and atomic resolvability. As a baseline benchmark, a 3D convolutional neural network trained on a curated, class-balanced, primarily high-resolution subset distinguished five motif/background classes, achieving macro-averaged sensitivity of 0.836 ± 0.019, specificity of 0.958 ± 0.005, balanced accuracy of 0.897 ± 0.012, and G-mean of 0.894 ± 0.013. AVAILABILITY AND IMPLEMENTATION: Source code, pipeline implementation, benchmark datasets, and an interactive web application are available at GitHub (https://github.com/DrDongSi/3DEM-RNA-Motif-Dataset), Zenodo (https://zenodo.org/communities/3dem-rna-motif-dataset), and Hugging Face Spaces (https://huggingface.co/spaces/houlab/arsma-cryoem).","[""Murugadass Chandramathi"", ""Rana Hajira"", ""Znosko Brent M"", ""Hou Jie"", ""Si Dong""]",Computational biology and chemistry,2026-08-30,10.1016/j.compbiolchem.2026.109344,"[""Journal Article""]","[""Cryoelectron Microscopy"", ""Machine Learning"", ""RNA"", ""Models, Molecular"", ""Nucleic Acid Conformation"", ""Benchmarking""]",PMC13631633 42685560,https://pubmed.ncbi.nlm.nih.gov/42685560/,Comparative machine learning-assisted terahertz and ultraviolet-visible spectroscopy for rapid authentication and quantitative assessment of adulteration levels in Chinese medicinal herbs.,"Reliable authentication and quantitative assessment of adulteration levels in traditional Chinese medicinal herbs are essential yet remain analytically challenging due to their complex chemical profiles, morphological similarities, and frequent adulteration. Here, we comparatively evaluate ultraviolet-visible (UV-Vis) spectroscopy, terahertz time-domain spectroscopy (THz-TDS), and their feature-level fusion combined with machine learning for rapid herbal authentication and adulteration-level prediction. Using Coptis chinensis as a representative target, the method was applied to three key analytical tasks: species identification, adulterant-type discrimination and regression-based adulteration-level prediction. Classification models were used for species identification and adulterant-type discrimination, while regression models were used to predict adulteration levels as continuous variables. After spectral preprocessing, supervised machine-learning models achieved high classification accuracies, exceeding 90% for THz-TDS data and reaching 100% for UV-Vis data. The regression analysis demonstrated proof-of-concept evaluation of adulteration-level prediction within the five predefined mixture levels, with lower internal cross-validation errors obtained from the UV-Vis spectra. Paired THz-TDS and UV-Vis spectra from the same samples were further used to directly compare THz-TDS alone, UV-Vis alone, and feature-level fusion. Under the standard methanol-extraction condition, feature fusion matched the classification performance of UV-Vis alone and showed generally comparable regression performance, while providing improved performance compared with THz-TDS alone. These results provide a comparative proof-of-concept assessment of the three spectral input strategies and establish a basis for future validation using broader sample sources and independent external mixtures.","[""Jin Yaxuan"", ""Xu Jia"", ""Jiang Xianguo"", ""Zhang Shaohua"", ""Hao Lei"", ""Yang Ning"", ""Meng Hao"", ""Zhou Chao"", ""Huang Wendeng"", ""Zhou Yan""]",Talanta,2026-08-29,10.1016/j.talanta.2026.130529,"[""Journal Article"", ""Comparative Study""]","[""Spectrophotometry, Ultraviolet"", ""Drugs, Chinese Herbal"", ""Machine Learning"", ""Terahertz Spectroscopy"", ""Drug Contamination"", ""Coptis"", ""Classification Algorithms""]", 42673671,https://pubmed.ncbi.nlm.nih.gov/42673671/,Hyperbolic graph contrastive learning for drug repositioning over heterogeneous biological networks.,"Drug repositioning has become an important computational strategy for identifying new therapeutic uses of existing drugs, especially when conventional drug discovery remains costly and time-consuming. However, accurate drug-disease association prediction remains challenged by the sparsity of verified associations and by the limited ability of Euclidean models to capture the hierarchical structure of heterogeneous biological networks. To address these issues, we propose HGCLDR, a hyperbolic graph contrastive learning framework for drug repositioning. HGCLDR first enhances the sparse drug-disease bipartite graph by incorporating similarity-derived structural information. It then introduces a protein-mediated two-hop topological projection module to infer biologically meaningful high-order drug-disease relations from drug-protein and protein-disease associations, while an adaptive denoising strategy is used to reduce hub-driven and propagation-induced noise. In addition, a layered sampling strategy is designed to construct semantically consistent yet structurally diverse contrastive views from observed associations and projected relations. Based on these views, drug and disease nodes are embedded on the Lorentz manifold and optimized through a hyperbolic graph contrastive learning framework, enabling the model to better capture the hierarchical and non-Euclidean characteristics of biological networks. Across the three benchmark datasets, HGCLDR attains the highest AUROC, AUPR, Accuracy, and Recall values; it also obtains the highest F1-score on the B- and C-datasets and a comparable F1-score on the F-dataset. Further evidence from ablation studies, case analyses, molecular docking, and a representative molecular dynamics simulation supports the effectiveness and biological relevance of the proposed framework.","[""Lu Pengli"", ""Ge Yu"", ""Wan Shengfang"", ""Peng Erjia"", ""Zhang Jun"", ""Yang Zhiwei""]",Computational biology and chemistry,2026-08-29,10.1016/j.compbiolchem.2026.109340,"[""Journal Article""]","[""Drug Repositioning"", ""Machine Learning"", ""Humans"", ""Algorithms""]", 42664649,https://pubmed.ncbi.nlm.nih.gov/42664649/,A GC-MS-based exhaled VOC diagnostic model for distinguishing early-stage lung cancer from controls.,"Lung cancer is a leading cause of cancer-related mortality, yet non-invasive tools for early-stage detection remain inadequate. Despite advances in breathomics, clinical translation is hindered by high-dimensional data complexity, algorithmic overfitting, and a lack of systematic exploration into subtle sub-stage differences such as Stage IA versus IB. This study aimed to identify exhaled volatile organic compounds (VOCs) biomarkers and develop diagnostic models for stage I non-small cell lung cancer (NSCLC), with exploratory subgroup analyses designed to preliminarily evaluate potential stage-specific heterogeneity. Breath samples from 74 patients with stage I NSCLC (52 IA, 22 IB) and 75 healthy controls were analyzed using thermal desorption gas chromatography-mass spectrometry (TD-GC-MS). A modular analytical framework combining multivariate statistics (PCA, OPLS-DA) and machine learning-based feature selection (LASSO, Random Forest, and SVM-RFE) was used to screen 562 VOCs. Trend and network analyses were performed to characterize metabolic shifts during disease progression. Diagnostic models were validated using nested cross-validation and a held-out independent test set. Concise VOCs signatures associated with stage I NSCLC and its subgroups were identified. Nineteen VOCs exhibited monotonic changes during progression, with 1,2-dichloroethane and cyclohexane identified as network-central VOCs. An RBF-kernel SVM model achieved an area under the curve (AUC) of 0.991 (0.956-1.000), with 93.3% (95% CI: 70.2-98.8%) sensitivity and 93.3% (95% CI: 70.2-98.8%) specificity in the test set. Additionally, a simplified two-VOC logistic regression model (cyclohexane and diethylsilane) yielded an AUC of 0.708. This study identifies stage-specific VOC biomarkers and demonstrates their potential for non-invasive detection. The exploratory subgroup analyses suggest possible biological heterogeneity within early-stage disease, and the parsimonious two-marker panel provides a practical foundation for developing low-cost, point-of-care screening strategies. These findings provide insights into early metabolic alterations and support the development of scalable breath-based screening strategies. Given the single-center design, these findings should be considered hypothesis-generating, with external validation in multicenter cohorts needed to determine their generalizability and clinical utility.","[""Wang Zhiying"", ""Liu Tianqing"", ""Liu Bohao"", ""Zhao Heng"", ""Huang Jiaqi"", ""Dong Huanhuan"", ""Wu Weiwei"", ""Zhang Guangjian""]",Talanta,2026-08-19,10.1016/j.talanta.2026.130485,"[""Journal Article""]","[""Humans"", ""Volatile Organic Compounds"", ""Lung Neoplasms"", ""Gas Chromatography-Mass Spectrometry"", ""Female"", ""Breath Tests"", ""Male"", ""Neoplasm Staging"", ""Middle Aged"", ""Carcinoma, Non-Small-Cell Lung"", ""Biomarkers, Tumor"", ""Aged"", ""Early Detection of Cancer"", ""Case-Control Studies""]", 42659825,https://pubmed.ncbi.nlm.nih.gov/42659825/,Analytical advances in citrus disease detection: Conventional molecular assays and emerging analytical technologies.,"Citrus diseases caused by fungal, bacterial, and viral pathogens continue to threaten global citrus production, creating an urgent need for analytical technologies that enable accurate, rapid, and field-deployable disease diagnosis. This review critically examines the evolution of citrus disease diagnostics, encompassing conventional molecular and immunological methods together with emerging electrochemical, spectroscopic, imaging, and artificial intelligence (AI)-assisted analytical approaches. Particular emphasis is placed on analytical performance, including sensitivity, selectivity, reproducibility, matrix effects, analytical robustness, and practical applicability, as well as on the contribution of nanomaterials and advanced sensing interfaces to improving analytical capability. Conventional techniques, including polymerase chain reaction (PCR), quantitative PCR (qPCR), enzyme-linked immunosorbent assay (ELISA), and lateral flow assays (LFA), are discussed alongside recent advances in electrochemical biosensors, Raman spectroscopy, surface-enhanced Raman spectroscopy (SERS), laser-induced breakdown spectroscopy (LIBS), hyperspectral imaging (HSI), fluorescence-based methods, and remote sensing technologies. The review further examines the growing role of chemometrics, machine learning, and deep learning in extracting diagnostic information from complex multidimensional datasets. It discusses their integration with Internet of Things (IoT) architectures and precision agriculture for continuous disease surveillance and decision support. Beyond analytical performance, critical challenges associated with technology readiness, standardization, scalability, commercialization, regulatory acceptance, external validation, and field implementation are comprehensively discussed. Overall, the evidence indicates that future citrus disease diagnostics will rely on interoperable and complementary analytical workflows that integrate molecular specificity with electrochemical, spectroscopic, imaging, and computational technologies to enable early disease detection, precision phytosanitary management, and robust decision-making under real agricultural conditions.","[""Silva Lucas Moreira"", ""Ferreira Antonio Aparecido Pupim"", ""Yamanaka Hideko""]",Talanta,2026-08-25,10.1016/j.talanta.2026.130509,"[""Journal Article"", ""Review""]","[""Citrus"", ""Plant Diseases"", ""Biosensing Techniques"", ""Artificial Intelligence"", ""Spectrum Analysis, Raman"", ""Electrochemical Techniques"", ""Enzyme-Linked Immunosorbent Assay""]", 42636660,https://pubmed.ncbi.nlm.nih.gov/42636660/,"""Gyroscope""-type portable device based on a bimetallic Fe/Zn-ZIF8 MOF nanozyme@agarose hydrogel with deep learning for monitoring Hg pollution.","Mercury (Hg) pollution in Hg mine waste residues under precipitation leaching causes long-term pollution to the surrounding environment, posing a serious threat to environmental and public health. To achieve rapid monitoring of Hg pollution in mining areas, a system (Fe/Zn-ZIF8@agarose hydrogel) was developed in this study using an agarose hydrogel as a bimetallic Fe/Zn-ZIF8 metal-organic framework (MOF) nanozyme carrier. This system leverages the specific recognition of Hg2+ by the thiol groups in the antioxidant glutathione (GSH) to regulate the peroxidase-like catalytic activity of Fe/Zn-ZIF8 in the H2O2 and 3,3',5,5'-tetramethylbenzidine (TMB) colorimetric reaction, enabling an ""on-off-on"" signal response. To further integrate deep learning-assisted image analysis algorithms, the sensing unit was embedded into a 3D-printed ""gyroscopic"" device to develop a portable platform for the visual detection of Hg2+ in water environments. This portable device exhibited excellent reliability for Hg2+ detection, with a spiked recovery rate of 95.8∼112.8% and a relative standard deviation (RSD) < 5.72%. It was successfully applied for onsite Hg2+ detection in mine leachate, surrounding river water, and paddy soil water. Among these samples, the Fe/Zn-ZIF8@agarose hydrogel was used to detect an Hg2+ concentration of 5.3 μM in mine leachate, whereas Fe/Zn-ZIF8 was used to detect Hg2+ concentrations of 0.36 μM and 0.12 μM in certain river water and paddy soil samples from mining areas, respectively. Through multimodule collaboration, this system significantly reduces the subjective errors associated with traditional colorimetric methods and provides a rapid, accurate, and user-friendly portable analytical platform for the onsite emergency monitoring of mercury pollution in mining areas.","[""Ma Wei"", ""Zhai Kun"", ""Mao Kang"", ""Yin Shupei"", ""Su Junxia"", ""Chen Chaofa"", ""Xiang Dongshan"", ""Du Wei"", ""Zhang Hua""]",Talanta,2026-08-19,10.1016/j.talanta.2026.130478,"[""Journal Article""]","[""Mercury"", ""Sepharose"", ""Zinc"", ""Hydrogels"", ""Metal-Organic Frameworks"", ""Deep Learning"", ""Environmental Monitoring"", ""Iron"", ""Colorimetry"", ""Water Pollutants, Chemical"", ""Hydrogen Peroxide"", ""Benzidines"", ""Glutathione""]", 42617305,https://pubmed.ncbi.nlm.nih.gov/42617305/,"Microplastic quantification and chemical characterization in salt samples, using stereomicroscopy, smartphone camera and supervised machine learning tools.","The microplastic (MPs) pollution, especially in the oceans, has a direct impact on seafood products, particularly sea salt. This is a growing concern due to the high daily consumption of salt and the potential adverse health effects associated with MPs exposure. This study presents a rapid, cost-effective, and green approach for MPs-detection in food samples employing stereomicroscopy, smartphone-cameras, image analysis, and Machine-Learning tools. The method integrates accessible image acquisition with open-source software (Ilastik/FIJI) to enable automated detection and measurement of two common morphologies: filaments and fragments, providing a straightforward alternative for laboratories with limited resources. The developed models demonstrated high sensitivity (100% filaments, 76.5% fragments) and acceptable specificity (76.4% filaments, 72.2% fragments), with an average recovery of 108.6% for spiked samples. Upon application to sea salt samples from México, the model identified 1950 ± 354 MPs pieces/kg of salt. Among the detected MPs, filaments were the predominant morphology (1235 ± 263 MPs pieces/kg), while fragment count was 715 ± 167 MPs pieces/kg. Polypropylene, polyethylene, Polyvinyl chloride, and cellophane were confirmed via FTIR analysis. The method was successfully applied to 6 additional food matrices (e.g., sugar, meat tenderizer). The proposed methodology is delimited to the detection of microplastics ≥300 μm (filaments) and ≥200 μm (fragments), and therefore does not account for smaller particles. The proposed method offers a rapid, replicable and accessible alternative that enables high-throughput preliminary microplastic detection while reserving chemical identification for complementary spectroscopic analysis, thus suggesting a potential for routine application in microplastic monitoring.","[""Hernández-Covarrubias Cristian M"", ""Ceballos-Magaña Silvia G"", ""Aguayo-Villarreal Ismael A"", ""Muñiz-Valencia Roberto""]",Talanta,2026-08-18,10.1016/j.talanta.2026.130467,"[""Journal Article""]","[""Microplastics"", ""Smartphone"", ""Microscopy"", ""Supervised Machine Learning"", ""Image Processing, Computer-Assisted""]", 42612250,https://pubmed.ncbi.nlm.nih.gov/42612250/,Machine learning-driven SERS quantification of phosphates through MB-Mo-AgNPs spectral transduction.,"An indirect surface-enhanced Raman spectroscopy (SERS) approach based on an MB-Mo-AgNPs probe system was developed to quantify aqueous KH2PO4 standard solutions. Silver nanoparticles (AgNPs), methylene blue (MB), and ammonium molybdate served as the SERS substrate, Raman reporter, and phosphate-responsive reagent, respectively. All concentrations were expressed as KH2PO4 mass concentrations. Across 0.0625-10 mg L-1 KH2PO4, the MB bands near 1400, 1470, and 1626 cm-1 decreased with increasing concentration. UV-Vis spectroscopy, qualitative dried-state transmission electron microscopy, dynamic light scattering, and zeta-potential measurements supported changes in the probe's colloidal state, particle association, and interfacial environment, consistent with a possible indirect signal-modulation pathway. These measurements did not establish a specific molecular mechanism or demonstrate MB desorption. The dataset comprised 2040 spectra from 51 independently prepared physical samples at 17 concentrations. In physical-sample-level grouped training-validation-test rotations, the 40 replicate spectra per sample remained within one subset, preventing sample overlap. Single-peak linear regression was compared with full-spectrum partial least squares regression, support vector regression (SVR), a conventional one-dimensional convolutional neural network (1D-CNN), and an attention-enhanced 1D-CNN. The full-spectrum models yielded lower errors than single-peak linear calibration. SVR achieved the lowest quantitative error among the compared models, with an RMSE of 0.0072 ± 0.0010 mg L-1 across three grouped rotations. In the prespecified grouped analysis, the attention-enhanced 1D-CNN yielded lower quantitative error than the conventional 1D-CNN. The results apply only to aqueous KH2PO4 standard solutions and do not validate total-phosphorus determination, real-wastewater analysis, or performance in external matrices.","[""Fu Yuhan"", ""Wang Shu"", ""Hu Jihui"", ""Zhou Yadong"", ""Zhan Chunlian"", ""Yin Lin"", ""Meng Yan Long""]",Talanta,2026-08-17,10.1016/j.talanta.2026.130414,"[""Journal Article""]",, 42600475,https://pubmed.ncbi.nlm.nih.gov/42600475/,Progresses in viral detection technologies for wastewater-based epidemiology.,"Wastewater-based epidemiology (WBE) has emerged as a powerful surveillance tool used to assess public health by tracking infectious materials shed into sewage systems by infected hosts. To enable a widespread implementation of WBE, advanced analytical technologies are required for timely, efficient, and accurate detection of the viruses from the complex water matrix. This review discusses key advancements in molecular methods and biosensors demonstrated on wastewater samples that largely enhance wastewater surveillance, including improved sensitivity, enhanced portability, shortened turnover time, integration with wastewater sample treatment, multiplexing ability, and the incorporation of artificial intelligence (AI)/machine learning (ML). We also compare the analytical performance of these advanced methods, their advantages, current challenges, and future direction on how they can better enable wastewater surveillance via efficient detection of targeted and emerging/untargeted viruses from wastewater.","[""Tanaka Yuki"", ""Salleh Nur Asinah Binte Mohamed"", ""Ow Sian Yang"", ""Zheng Xin Ting"", ""Su Xiaodi"", ""Sutarlie Laura""]",Talanta,2026-08-08,10.1016/j.talanta.2026.130419,"[""Journal Article"", ""Review""]","[""Viruses"", ""Biosensing Techniques"", ""Wastewater"", ""Wastewater-Based Epidemiological Monitoring"", ""Humans"", ""Machine Learning"", ""Artificial Intelligence"", ""Environmental Monitoring""]", 42594442,https://pubmed.ncbi.nlm.nih.gov/42594442/,Machine learning-assisted universal PGM sensing platform: Recognition-detection separation via DNAzyme-nanozyme triplex switch.,"Constructing a universal, accurate personal glucose meter (PGM)-based detection strategy via separating recognition and detection to eliminate matrix interference, and a versatile signal transduction method without glycosidases, can enhance PGM detection universality and accuracy, expanding its point-of-care testing (POCT) applications beyond blood glucose detection. Herein, we develop a highly sensitive PGM-based biosensor for ochratoxin A (OTA) detection by integrating a triplex molecular switch (THMS) with DNAzyme-mediated signal amplification using an Au@Pt nanozyme. The THMS, assembled from an aptamer and a Pb2+-dependent DNAzyme, specifically recognizes OTA, triggering THMS dissociation and DNAzyme release. The liberated DNAzyme cleaves the Au@Pt-SingleDNA-MB signal tag, releasing Au@Pt nanoparticles for PGM readout via glucose oxidation. Conventional four-parameter logistic (4 PL) regression was used to establish the calibration curve, which showed a linear range of 0-0.05 ng/mL and a detection limit of 0.008 ng/mL; this detection limit is 10-fold lower than that of Au NPs-based methods. Additionally, machine learning was employed to further analyze the performance of the biosensor. Recovery experiments in corn and wine samples gave acceptable recoveries of 89.09%-116.66% and 83.90%-109.09%, demonstrating the practical applicability of the biosensor. Overall, this study provides a portable and sensitive POCT platform for the on-site detection of mycotoxins.","[""Dong Guannan"", ""Chai Liting"", ""Jian Mingying"", ""Niu Jianning"", ""Feng Zihan"", ""Liang Jiayu"", ""Wang Yanru"", ""Wang Jianlong"", ""Ma Min""]",Talanta,2026-08-11,10.1016/j.talanta.2026.130432,"[""Journal Article""]","[""DNA, Catalytic"", ""Machine Learning"", ""Biosensing Techniques"", ""Ochratoxins"", ""Gold"", ""Metal Nanoparticles"", ""Limit of Detection"", ""Blood Glucose Self-Monitoring"", ""Aptamers, Nucleotide""]", 42594440,https://pubmed.ncbi.nlm.nih.gov/42594440/,"Recent advances in lateral flow immunoassay for virus detection: current achievements, challenges and future prospects.","Infectious diseases continue to pose significant global public health challenges, highlighting the urgent need for rapid, accessible, and scalable diagnostic strategies. Lateral flow immunoassay (LFIA), characterized by its rapid response, low cost, and instrument-free operation, has been widely adopted for infectious disease detection. However, it is still predominantly regarded as a point-of-care testing (POCT) tool, while its broader role in public health systems remains underappreciated. In this review, LFIA is repositioned as a point-of-need (PoN) self-testing platform with the potential to enable large-scale screening, which is critical for epidemic surveillance and pandemic preparedness. We systematically summarize recent advances in LFIA from three key perspectives: (i) signal generation and transduction strategies based on diverse nanomaterials, including colorimetric, fluorescent, magnetic, SERS, and photothermal systems; (ii) multimodal and multiplex detection approaches for improving sensitivity, accuracy, and throughput; and (iii) the integration of machine learning (ML) for signal interpretation and system optimization. In addition, we comprehensively review the applications of LFIA in detecting viral antigens, antibodies, and nucleic acids. Finally, the current limitations of LFIA, including sensitivity constraints, target specificity, and practical challenges related to cost, usability, and stability, are critically discussed. Future development directions are proposed, emphasizing advanced nanomaterials, intelligent data processing, and multi-target detection strategies. Overall, this review highlights the transformative potential of LFIA, particularly colorimetric LFIA, as an emergency pandemic screening tool, as well as its growing significance in routine decentralized testing and strengthening global public health preparedness.","[""Li Linzhi"", ""Liu Fang"", ""He Xiran"", ""Yuan Dan"", ""Hao Senlin"", ""Wang Lexin"", ""Li Xiaosong"", ""Mao Binli""]",Talanta,2026-08-05,10.1016/j.talanta.2026.130402,"[""Journal Article"", ""Review""]","[""Immunoassay"", ""Humans"", ""Rapid Diagnostic Tests"", ""Point-of-Care Testing"", ""Viruses"", ""COVID-19""]", 42580250,https://pubmed.ncbi.nlm.nih.gov/42580250/,Non-destructive biochemical profiling of goldenberry ripening using confocal micro Raman spectroscopy coupled with machine learning.,"The accurate monitoring of the fruit ripening is crucial for maintaining quality and prevention from post-harvest economic losses. Goldenberry, being rich in vitamins, minerals, and carotenoids, yet its ripening assessment relies on variable and often subjective traditional approaches. To address this, the present study utilises rapid and non-destructive techniques using confocal micro-Raman and UV-Vis spectroscopy to track biochemical changes in both the exocarp and mesocarp of goldenberry fruit across five ripening stages: pre-mature, mature, pre-ripe, ripe, and post-ripe. The analysis of Raman spectra of exocarp shows an increase in carotenoid bands up to ripe stage, followed by a decrease at post-ripe stage, while the mesocarp exhibits a continuous increase in carotenoid intensity through stage post-ripe. The UV-Vis spectral analysis supports these observations, indicating increasing carotenoid content alongside a reduction in chlorophyll as ripening progresses. Among different machine learning (ML) approaches, unsupervised clustering demonstrates strong performance of k-means and agglomerative nesting, with high rand index values (>0.97). In supervised domain, support vector machine, logistic regression, random forest, decision tree, and principal component-linear discriminant analysis achieves the highest values of performance metrices including accuracy, precision, recall, F1 score, ROC/AUC values, and cross validation scores. Moreover, partial least squares regression analysis demonstrates excellent predictive power, with R2 = 0.93 (calibration) and 0.70 (cross-validation) for exocarp and R2 = 0.91 (calibration) and 0.89 (cross-validation) for mesocarp. The combined use of Raman with ML enables a practical and reliable framework for assessing fruit ripening, with promising applications in quality control and post-harvest management.","[""Awasthi Aishwary"", ""Tripathi Aradhana"", ""Sharma Shristi"", ""Tiwari Aparna"", ""Baran Chhavi"", ""Sharma Sweta"", ""Uttam K N""]",Talanta,2026-08-05,10.1016/j.talanta.2026.130388,"[""Journal Article""]","[""Spectrum Analysis, Raman"", ""Fruit"", ""Machine Learning"", ""Carotenoids"", ""Random Forest"", ""Support Vector Machine""]", 42561794,https://pubmed.ncbi.nlm.nih.gov/42561794/,Decoupling human hair isotope codes via mechanistic feature construction and calibrated evidence fusion for uncertainty-aware forensic attribution.,"Stable isotope analysis of human hair (δ2H, δ18O, δ13C, and δ15N) offers a non-invasive trajectory for geographic sourcing, yet its forensic application remains constrained by overlapping isotopic catchments, individual lifestyle variations, and the high risk of overinterpretation in deterministic machine-learning classifiers. Here, we present a chemically interpretable and uncertainty-aware framework that transitions hair isotope forensics from empirical pattern recognition to rigorous evidence quantification. To impose a chemistry-informed representation on the correlated isotope measurements, we constructed the Resident-Hair Water-Diet Isotope Descriptor (RH-WDID) system by integrating hydrological gradients, dietary contrasts, and cross-domain coupling features. Crucially, all mathematical transformations were constrained strictly within training folds to eliminate information leakage. Compared with the raw isotope space, the RH-WDID feature space markedly enhanced spatial compactification and separability, increasing the unsupervised silhouette coefficient from 0.178 to 0.305 and reducing the Davies-Bouldin index from 1.517 to 1.325. Systematic benchmarking identified a residual tabular multilayer perceptron (ResidualTabMLP) as the optimal discrimination backbone, achieving a high internal five-fold cross-validated accuracy of 0.937, with mechanistic validity supported by SHAP analysis and forward/reverse ablation experiments. Importantly, this value represents internal cross-validation performance only; no independent external test cohort collected from another region, collection period, or laboratory was available in the present study. Therefore, the model should be interpreted as a first-stage, scope-defined seven-city resident-reference framework rather than as a fully externally validated forensic classifier. To reduce the risk of overinterpretation, this deterministic backbone was further upgraded into a Resident-Hair Forensic Isotope Evidence-Fusion (RH-FCIEF) model. Instead of forced-choice top-1 labels, RH-FCIEF outputs a calibrated multi-model probability vector coupled with scope-aware typicality assessments. This study provides a chemically interpretable and uncertainty-aware modelling route for converting bulk isotope measurements into probabilistic resident-source evidence within a clearly defined reference scope.","[""Jie Zhaowei"", ""Zhu Jun"", ""Tao Qing"", ""Zheng Hanyang"", ""Cai Zhuotong"", ""Hu Can"", ""Li Yajun"", ""Guo Hongling"", ""Mei Hongcheng""]",Talanta,2026-08-03,10.1016/j.talanta.2026.130395,"[""Journal Article""]","[""Humans"", ""Hair"", ""Carbon Isotopes"", ""Nitrogen Isotopes"", ""Uncertainty"", ""Forensic Sciences"", ""Oxygen Isotopes"", ""Deuterium"", ""Calibration"", ""Isotopes"", ""Multilayer Perceptrons""]", 42541882,https://pubmed.ncbi.nlm.nih.gov/42541882/,Machine learning-assisted colorimetric sensor array on a superwettable microchip for on-site authentication of ginseng.,"Panax ginseng is widely used for its significant bioactivities, but adulteration and quality counterfeiting remain critical issues. Herein, a machine learning-assisted superwettable colorimetric sensor array microchip is proposed for the rapid ginseng authentication. The microchip targets diol groups of ginseng-derived monosaccharides using 3-nitrophenylboronic acid as the cross-reactive receptor, generating distinct colorimetric fingerprints via pH response and indicator displacement assay (IDA) mechanisms. The excellent determination capability of the microchip was demonstrated through the classification and semi-quantitative analysis of monosaccharides in ginseng hydrolysates. Coupled with advanced machine learning algorithms, high-accuracy analysis was achieved for ginseng species identification and authenticity classification, reaching 96.49% and 99.02%, respectively. Furthermore, support vector regression (SVR) enabled precise quantitative analysis of adulteration ratios with excellent predictive performance (R = 0.9869). The proposed method was successfully applied to identify commercial ginseng samples with reliable results. This work provides a low-cost, rapid, and instrument-free strategy for on-site quality control of ginseng products.","[""Jiang Huan"", ""Ma Bingxin"", ""Wang Xuyang"", ""Yang Yuemeng"", ""Feng Zihui"", ""Liu Li"", ""Tang Xiaolei"", ""Xu Li-Ping""]",Talanta,2026-08-01,10.1016/j.talanta.2026.130385,"[""Journal Article""]","[""Panax"", ""Colorimetry"", ""Machine Learning"", ""Support Vector Machine""]", 42508751,https://pubmed.ncbi.nlm.nih.gov/42508751/,Automated detection of Eimeria tenella from hematoxylin and eosin-stained chicken cecal tissues using YOLOv4-based deep learning: A proof-of-concept study.,"Avian coccidiosis is a disease primarily characterized by diarrhea caused by intestinal infection with protozoa of the genus Eimeria, causing significant economic impact worldwide. Early diagnosis of infected chickens is crucial for effective control of this disease. While several studies have reported the usefulness of deep learning algorithms for recognizing oocysts on smear specimens, to the authors' knowledge, no reports have examined this approach on histopathological specimens. In this study, based on the hypothesis that deep learning can assist histopathological diagnosis, we applied an object detection algorithm to demonstrate models to automatically detect schizonts and macrogametocytes of E. tenella from histopathological images as a proof-of-concept. We prepared 960 image patches from hematoxylin and eosin-stained cecal specimens of four chickens orally infected with E. tenella. These were divided into training and validation data, and models were developed using the You Only Look Once version 4 (YOLOv4) and YOLOv4‑tiny algorithms. The resulting models were then used to detect parasites in tissue images from field cases. The resulting mean average precisions were 78.08-78.87% for YOLOv4 and 64.26-68.13% for YOLOv4‑tiny. Although this study has limitations in the dataset size and may not fully capture the real-world variability, it suggests that YOLOv4 is feasible for detecting coccidia in histopathological specimens. The proposed approach has the potential to support histopathological diagnosis and facilitate the detection of parasitic stages during routine pathological examinations, contributing to more objective and reproducible pathological evaluations.","[""Iwaide Susumu"", ""Kimura Kumiko"", ""Matsubayashi Makoto"", ""Murakami Tomoaki"", ""Shibahara Tomoyuki""]",Parasitology international,2026-07-27,10.1016/j.parint.2026.103353,"[""Journal Article""]","[""Animals"", ""Chickens"", ""Cecum"", ""Eimeria tenella"", ""Coccidiosis"", ""Poultry Diseases"", ""Deep Learning"", ""Proof of Concept Study"", ""Hematoxylin"", ""Eosine Yellowish-(YS)"", ""Detection Algorithms"", ""Staining and Labeling""]", 42475951,https://pubmed.ncbi.nlm.nih.gov/42475951/,Intelligent nanozyme-based colorimetric sensor array for identifying six major teas and quantifying adulteration.,"The rapid discrimination of tea varieties and detection of adulteration are critical for food safety and quality control. In this study, a three-channel colorimetric sensor array was constructed using MIL100(Fe), Au nanoclusters (Au NCs), and MIL100(Fe)/Au NCs to distinguish tea polyphenols. The sensing mechanism is based on the differential inhibition of peroxidase-like activity arising from the interaction between structurally diverse tea polyphenols and the active sites, which enables the establishment of specific characteristic fingerprint patterns for each analyte. The experimental results demonstrated that, in the current dataset, the colorimetric sensor array reached a maximum internal cross-validation accuracy of 100% for five tea polyphenols and exhibited excellent clustering across a dynamic range of 10 nM to 10 mM. Through a systematic evaluation of 8 machine learning algorithms and the optimization of the K-Nearest Neighbors (KNN) model by combining data fusion strategies, the highest internal cross-validation accuracy for classifying six major teas reached 100%. Furthermore, a linear quantitative identification model for Longjing tea adulteration (ranging from 0% to 100%) was established, along with nanomolar-level ultrasensitive detection of specific tea polyphenols. This study confirms that machine learning-assisted nanozyme sensor arrays has significant application potential in the rapid assessment of tea quality and market supervision.","[""Wang Xiaoyan"", ""Zhao An"", ""Cao Zhen"", ""Lv Jingfei"", ""Lin Siyu"", ""Fu Xiaoying"", ""Liu Wei""]",Talanta,2026-07-20,10.1016/j.talanta.2026.130322,"[""Journal Article""]","[""Tea"", ""Colorimetry"", ""Polyphenols"", ""Metal Nanoparticles"", ""Machine Learning"", ""Food Contamination"", ""Gold"", ""Biosensing Techniques""]", 42472529,https://pubmed.ncbi.nlm.nih.gov/42472529/,Hierarchically ordered ERGO-L-Cys flower-like nanospheres on screen-printed electrodes for electrochemical immunosensing of cow's milk specific IgE with machine learning-assisted calibration.,"Cow's milk-specific IgE (sIgE) is a crucial serum biomarker for cow's milk allergy (CMA) in children. However, its low abundance and the scarcity of reference standards have made it challenging to develop effective electrochemical detection platforms for clinical use. This study reports, to the best of our knowledge, the first electrochemical immunosensing platform for the quantitative detection of cow's milk sIgE (CM-sIgE) in pediatric serum. Through the l-cysteine (L-Cys)-mediated electrochemical reduction of graphene oxide, a hierarchically ordered three-dimensional electrochemically reduced graphene oxide functionalized with L-Cys (ERGO-L-Cys) flower-like nanosphere interface was in situ constructed on a screen-printed carbon electrode, distinguishing it from conventional two-dimensional ERGO films. Combined with a biotin-streptavidin bridged alkaline phosphatase/1-naphthyl phosphate (ALP/1-NPP) enzyme amplification system, the quantitative detection of serum sIgE was achieved using differential pulse voltammetry. This platform exhibits a linear range of 0.56-54.52 IU/mL for serum sIgE, with a detection limit of 0.29 IU/mL, demonstrating good selectivity, reproducibility, and storage stability. To overcome the challenge of the lack of sIgE reference standards, pediatric serum samples calibrated against hospital chemiluminescence assay values were utilized. Moreover, random forest (RF) regression was applied for data-driven calibration, further enhancing the quantitative robustness in complex serum matrices. The analysis of clinical samples showed good agreement with the results from hospital chemiluminescence assays. This work provides an eco-friendly, reproducible approach for constructing three-dimensional biofunctionalized interfaces and offers a feasible solution for point-of-care electrochemical detection of CMA in children.","[""Yang Meiqing"", ""Xie Congkai"", ""Gao Xiaofei"", ""Jiang Xinjie"", ""Yang Jifei"", ""Lu Haozi"", ""Zhu Xingyuan"", ""Liu Song""]",Talanta,2026-07-18,10.1016/j.talanta.2026.130333,"[""Journal Article""]","[""Immunoglobulin E"", ""Animals"", ""Electrodes"", ""Electrochemical Techniques"", ""Humans"", ""Biosensing Techniques"", ""Graphite"", ""Cattle"", ""Nanospheres"", ""Machine Learning"", ""Milk Hypersensitivity"", ""Calibration"", ""Cysteine"", ""Milk"", ""Immunoassay"", ""Limit of Detection"", ""Reproducibility of Results""]", 42468169,https://pubmed.ncbi.nlm.nih.gov/42468169/,"MgFe2O4/graphene nanocomposites for bifunctional wastewater remediation: A critical review of adsorption, visible-light photocatalysis, and machine learning-assisted optimization.","MgFe2O4/graphene nanocomposites have emerged as promising bifunctional platforms combining magnetic adsorption and visible-light photocatalysis for wastewater remediation; yet no comprehensive quantitative review of their structure-performance relationships exist. This review critically evaluates MgFe2O4/graphene nanocomposites for simultaneous removal of heavy metal ions and synthetic organic dyes (2015-2025). A comparative evaluation of five synthesis methodologies are performed against four criteria that determine the performance (Brunauer-Emmett-Teller surface area (BET): 120-320 m2/g; Raman D/G ratio: 1.05-1.35) of the synthesized materials, and the hydrothermal and solvothermal methods are found to be optimal. This review proposes a preliminary first Pearson correlation framework for this material system has been supported by a compiled adsorption dataset (qmax: 35 to 930 mg/g). Statistical analysis determined that the sole factor significantly correlated with the photocatalytic rate constant is the identifying bandgap energy (Eg): r = -0.889, p = 0.003, n = 8. The best range of graphene loadings has been found to be between 5 and 10 wt%, in an empirical manner, because the required magnetic recovery (Ms > 10 emu/g) is met while being accompanied by the maximum adsorption capacity. A new transferable machine learning optimization protocol (TMLOP) is presented, which combines the elements of response surface methodology, artificial neural networks, and random forest feature importance. pH and initial concentration were identified as dominant operational predictors (combined importance >55%). Real wastewater matrix effects, regeneration mechanisms, nanoparticle leaching, and scale-up engineering gaps are critically assessed. Six evidence-based future research priorities are ranked by impact and feasibility, with MgFe2O4-specific ML modeling and Z-scheme heterostructure development identified as the highest priority directions.","[""Shareef Hafiza Hira"", ""Sadiq Huma"", ""Chaudhry Warda"", ""Shah Aiman"", ""Sharif Maha""]",Talanta,2026-07-12,10.1016/j.talanta.2026.130276,"[""Journal Article"", ""Review""]",, 42459054,https://pubmed.ncbi.nlm.nih.gov/42459054/,Deep Model Families for EEG-Based Multi-Class Dementia Classification.,"Deep learning (DL) has shown considerable promise for EEG-based dementia assessment; however, rigorous cross-family comparisons under leakage-free and clinically meaningful evaluation protocols remain limited. To address this gap, we benchmarked twelve DL architectures from four major families - recurrent neural networks (RNNs), temporal convolutional networks (TCNs), Transformers, and self-supervised learning (SSL) - for the multi-class differentiation of Alzheimer's disease (AD), frontotemporal dementia (FTD), and healthy controls (HC). The analysis was conducted on a clinically acquired EEG dataset comprising 36 AD patients, 23 FTD patients, and 29[Formula: see text]HC participants. To prevent subject-level data leakage, we employed strict subject-independent five-fold cross-validation, and to improve clinical relevance, final predictions were aggregated at the subject level using majority voting across EEG segments. Model performance was evaluated using classification metrics, learning stability, statistical ranking, computational efficiency, and interpretability analysis. Among all benchmarked architectures, the self-supervised SimCLR-Time framework achieved the strongest subject-level performance, reaching a mean accuracy of [Formula: see text] and a weighted F1-score of [Formula: see text]. The time-series transformer (TST) was the best-performing fully supervised model, with an accuracy of [Formula: see text], followed closely by the hybrid LSTM-FCN model at [Formula: see text]. In contrast, the EEG-Seq2Seq model showed the weakest performance ([Formula: see text]). Computational profiling further showed that TCN-Attention offered a favorable trade-off between accuracy and edge-oriented efficiency, while SimCLR-Time combined strong diagnostic performance with the fastest inference latency. In addition, Integrated Gradients-based saliency analysis suggested that the top-performing model relied on structured spatiotemporal EEG patterns rather than arbitrary temporal artifacts. Overall, these findings provide a controlled within-dataset benchmark for multi-class dementia classification from raw EEG and suggest that self-supervised and attention-based architectures are promising directions for clinically oriented EEG decision-support systems. However, external validation on larger and more heterogeneous cohorts remains necessary before broader generalization can be claimed.","[""Kotan Soner"", ""Akan Aydin""]",International journal of neural systems,2026-07-16,10.1142/S0129065727500055,"[""Journal Article""]","[""Humans"", ""Electroencephalography"", ""Deep Learning"", ""Alzheimer Disease"", ""Frontotemporal Dementia"", ""Convolutional Neural Networks"", ""Recurrent Neural Networks"", ""Classification Algorithms""]", 42636558,https://pubmed.ncbi.nlm.nih.gov/42636558/,Ethnoveterinary herbal practices are associated with reduced within-farm antibiotic use and improved farm performance in commercial poultry in Bangladesh: A multi-district cross-sectional study.,"ETHNOPHARMACOLOGICAL RELEVANCE: Antimicrobial resistance (AMR) in livestock is a critical One Health challenge, particularly in low- and middle-income countries (LMICs) where antibiotic regulation and surveillance are inadequate. Ethnoveterinary herbal practices are widely used in Bangladeshi commercial poultry farming but remain empirically undocumented at scale, limiting evidence-based AMR stewardship policy. AIM OF THE STUDY: To characterize ethnoveterinary herbal practices among commercial poultry farmers across 35 districts of Bangladesh, by documenting the plant species used, preparation methods, plant parts used, and target diseases, and to assess their observational associations with retrospective antibiotic use trajectories, farm performance outcomes, and food safety awareness. MATERIALS AND METHODS: A cross-sectional survey was conducted among 795 commercial poultry farmers (285 herbal users, 510 non-users) across 35 districts representing all 8 administrative divisions of Bangladesh (January-March 2026). The cohort included broiler, layer, and Sonali production systems. Statistical analyses included the Mann-Whitney U test (between-group performance comparisons), the Wilcoxon signed-rank test (within-group antibiotic shift), the chi-square test (categorical variables), multivariable logistic regression (predictors of herbal adoption), and multiple linear regression (predictors of antibiotic reduction magnitude). All analyses were conducted in Python 3.10 (scipy.stats v1.11, statsmodels v0.14; α = 0.05, two-tailed). Six machine learning classifiers with SHapley Additive exPlanations (SHAP) were applied as an exploratory investigation of potential predictors of farmer satisfaction. RESULTS: Among 795 surveyed farmers, the prevalence of herbal remedy use was 35.8% (n = 285). Fourteen medicinal plant species were documented; the most prevalent were Moringa oleifera Lam. (22.8%), Zingiber officinale Roscoe (21.4%), and Citrus × limon (L.) Osbeck (20.0%). Among herbal users (n = 285), self-reported antibiotic use decreased from a mean of 7.02 courses per cycle before adoption to 3.96 courses per cycle after adoption (43.6% within-group reduction; Wilcoxon W = 0, p < 0.001). This reflects a self-reported within-group temporal shift only; no between-group antibiotic comparison was possible. Herbal users had significantly lower Feed Conversion Ratio (FCR; 1.970 ± 0.261 vs. 2.144 ± 0.273; Mann-Whitney p < 0.001, r = 0.349) and mortality rate (3.42 ± 1.75% vs. 5.97 ± 2.30%; p < 0.001, r = 0.592) than non-users; these are observational, non-causal associations. The strongest independent predictors of herbal adoption were neighbor herbal use (OR = 2.741, 95% Bootstrap CI: 2.550-2.940, p < 0.001) and AMR knowledge items (OR = 1.511, 95% CI: 1.298-1.776, p < 0.001). All six machine learning classifiers showed near-random discriminative performance (AUC range: 0.407-0.535). Only 30.2% of herbal users were aware of potential phytochemical residue risks in meat and eggs. CONCLUSIONS: The findings provide a multi-district empirical baseline and an observational signal-suggestive but strictly non-causal-that ethnoveterinary herbal practices are associated with lower within-group antibiotic course frequencies and distinct production parameters. The prominent food safety awareness gap-fewer than one-third of herbal users recognizing residue risks-highlights an immediate need for targeted veterinary education, pharmacokinetic residue evaluation, and regulatory standardization of crude herbal veterinary inputs.","[""Nirjus Jubayer Mahmud"", ""Saha Mousumi"", ""Sarker Rudra""]",Journal of ethnopharmacology,2026-08-24,10.1016/j.jep.2026.122316,"[""Journal Article""]","[""Animals"", ""Bangladesh"", ""Cross-Sectional Studies"", ""Anti-Bacterial Agents"", ""Humans"", ""Animal Husbandry"", ""Plants, Medicinal"", ""Female"", ""Poultry"", ""Male"", ""Phytotherapy"", ""Chickens"", ""Plant Preparations"", ""Poultry Diseases""]", 42586397,https://pubmed.ncbi.nlm.nih.gov/42586397/,Stephania cepharantha Hayata (Baiyaozi) attenuates cirrhosis-associated intestinal barrier injury by restoring mitochondrial function and suppressing PI3K/AKT signaling.,"ETHNOPHARMACOLOGICAL RELEVANCE: Baiyaozi, the dried tuberous root of Stephania cepharantha Hayata, has long been used in traditional Chinese medicine for acute hepatitis, bacillary dysentery, acute gastroenteritis, abdominal pain, and bleeding disorders, indications that collectively correspond to hepatic inflammation, intestinal mucosal injury, and the bleeding tendency characteristic of cirrhosis-associated intestinal barrier injury. This ethnopharmacological correspondence provides a rationale for evaluating Baiyaozi (Stephania cepharantha Hayata) as a candidate therapy for the cirrhotic intestinal barrier, a setting in which current microbiota-targeted strategies act only indirectly on the intestinal epithelium and no traditional medicine has been shown to act directly on its cell-intrinsic driver, mitochondrial dysfunction. AIM OF THE STUDY: To identify mitochondria-related therapeutic targets for cirrhosis-associated intestinal barrier injury from human intestinal transcriptomic data, to identify Baiyaozi as a candidate medicine through reverse network pharmacology, and to validate its efficacy and underlying mechanism in vivo, focusing on the restoration of mitochondrial function and suppression of PI3K/AKT signaling. MATERIALS AND METHODS: We performed differential expression analysis on GSE285291 (human duodenal and ascending colon tissues; 28 cirrhotic patients, 8 healthy controls) and intersected DEGs with 2719 mitochondrial dysfunction-related genes (MDRGs) to identify MDRDEGs. A three-step machine learning pipeline (Random Forest, SVM-RFE, LASSO) constructed a diagnostic model. Hub genes were identified by PPI analysis. Candidate TCMs were predicted via Coremine and TCMSP, and binding affinities assessed by molecular docking. The chemical constituents of the Baiyaozi decoction were characterized by UHPLC-Q-TOF-MS/MS. A CCl4-induced murine cirrhosis model validated the intestinal protective effects of Baiyaozi. RESULTS: 177 MDRDEGs were identified. The five-gene diagnostic model (JAK2, NDUFA2, PIK3CA, PRKAA1, SHMT2) achieved AUC = 0.996. JAK2 and PIK3CA were hub genes. Among TCM candidates, Baiyaozi alone met pharmacokinetic screening thresholds, and its active alkaloids exhibited strong binding affinities to both JAK2 and PIK3CA (-6.7 to -10.5 kcal/mol). UHPLC-Q-TOF-MS/MS tentatively identified 14 alkaloids in the decoction, including cepharanthine, tetrandrine, and crebanine. Crebanine, also predicted in silico, showed the strongest JAK2 affinity (-9.9 kcal/mol), while cepharanthine and tetrandrine exhibited the highest PIK3CA affinities (-10.5 and -10.0 kcal/mol). In vivo, Baiyaozi attenuated colonic mucosal injury, restored tight junction proteins, rescued mitochondrial markers (COXIV, ATP5A, TFAM, NDUFS1), normalized ATP and mtDNA copy number, and suppressed aberrant p-PI3K/p-AKT activation. CONCLUSIONS: This study establishes the first mitochondria-centric diagnostic model for cirrhosis-associated intestinal mucosal injury and demonstrates that Baiyaozi exerts intestinal barrier-protective effects through concomitant restoration of mitochondrial energy homeostasis and suppression of PI3K/AKT signaling, a mechanism distinct from the microbiota-mediated pathways reported for other TCMs.","[""Zhou Yimeng"", ""Sun Xu"", ""Chang Di""]",Journal of ethnopharmacology,2026-08-12,10.1016/j.jep.2026.122308,"[""Journal Article""]","[""Animals"", ""Proto-Oncogene Proteins c-akt"", ""Humans"", ""Signal Transduction"", ""Mitochondria"", ""Phosphatidylinositol 3-Kinases"", ""Intestinal Barrier Function"", ""Male"", ""Drugs, Chinese Herbal"", ""Stephania"", ""Liver Cirrhosis"", ""Intestinal Mucosa"", ""Mice"", ""Mice, Inbred C57BL"", ""Network Pharmacology""]", 42913907,https://pubmed.ncbi.nlm.nih.gov/42913907/,Cognitive agent neuron: Learning and decision-making inspired by dopaminergic neurons.,"Decision-making is a critical function of the brain as an information-processing system that can be understood using mathematical models, computer simulations, and theoretical analysis. In this paper, we introduce a computational neuron model inspired by dopaminergic functionality, which we call Cognitive Agent Neurons (CANs). Dopaminergic neurons play an important role in transmitting global reward prediction errors (RPEs) across a wide array of brain regions. RPEs serve as a key signal in adaptive learning and behavior, directly impacting reinforcement learning mechanisms. We propose a rational neuron model as a neuronal agent that can communicate with its environment and improve its decision-making skills over time, inspired by dopaminergic neurons. The network of neuronal agents can cooperate to solve any task and improve their functionality with the reinforcement learning algorithm over time. So, the CAN model is an information processing system that is based on the premise that each neuron functions as an autonomous and rational agent. We examine the functionality of the CAN model based on the classification task. Due to the chemical synapse, we proposed two Dynamic (extensive-form) games with imperfect information to model the interaction of neurons in the CAN model. After training the model with approximate Q-learning, each neuron can choose the neuron that sends its signal. The results reveal that the CAN model's accuracy rate for dividing the banknote authentication dataset is 97%, its accuracy rate for dividing two classes of the notMNIST dataset is 92%, and its accuracy rate for dividing the non-linear dataset is 96%. After training the model with approximate Q-learning, the results show that the CAN model performs better in classification due to its extensive action space. It performs comparably to traditional machine learning methods.","[""Zareh Masoumeh"", ""Manshaei Mohammad Hossein"", ""Zahabi Sayed Jalal"", ""Saghali Mohammad"", ""Nematbakhsh Golnoush""]",Journal of theoretical biology,2026-10-22,10.1016/j.jtbi.2026.112628,"[""Journal Article""]","[""Dopaminergic Neurons"", ""Decision Making"", ""Models, Neurological"", ""Learning"", ""Cognition"", ""Reinforcement Machine Learning"", ""Humans"", ""Animals"", ""Computer Simulation"", ""Reward"", ""Algorithms""]", 42913359,https://pubmed.ncbi.nlm.nih.gov/42913359/,From detection to prediction: AI-integrated molecular and biosensing approaches for psychrotrophic and thermophilic spoilage bacteria in dairy systems.,"Psychrotrophic Gram-negative bacteria and thermophilic or thermoduric spore-forming bacteria are major causes of quality deterioration and shelf-life loss in milk and dairy products. Their impact is mediated primarily by heat-stable peptidases and lipases, heat-resistant spores and biofilms that persist in processing environments; with the notable exception of toxigenic members of the Bacillus cereus group, most of these organisms are spoilage rather than safety hazards. Culture-based methods remain the reference standard but are slow and provide single time-point information. This review critically evaluates molecular and sensing methods, including quantitative and digital PCR, isothermal amplification, CRISPR-Cas assays, biosensors, viability assays, spectroscopy and (meta)omics, with explicit attention to whether performance has been demonstrated in naturally contaminated dairy matrices rather than in buffer or spiked samples, and examines how their outputs could inform predictive microbiology and machine learning. A classification framework based on growth-temperature range, heat resistance, sporulation and dairy relevance is provided, and each detection-to-prediction pathway is graded as demonstrated in dairy, partly demonstrated or proposed. Validated dairy applications currently exist mainly for individual detection assays and for mechanistic or stochastic shelf-life models driven by culture-derived data; prospective, validated workflows in which molecular or biosensor outputs are used as model inputs to forecast spoilage have not yet been reported. Because heat-stable enzymes can spoil ultra-high-temperature products after the producing cells have been inactivated, viable counts alone are insufficient predictors of such defects. Key barriers include matrix effects, small and imbalanced datasets, domain shift, limited external validation and poorly defined intended use. Detection-to-prediction integration should therefore be regarded as an emerging research direction rather than an established industrial capability.","[""Shamlan Arzaq"", ""Khan Mati Ullah"", ""Farid Anum"", ""Huo Shengnan"", ""Liu Hong"", ""Chen Zhiwei""]",International journal of food microbiology,2026-10-21,10.1016/j.ijfoodmicro.2026.112130,"[""Journal Article"", ""Review""]","[""Biosensing Techniques"", ""Food Microbiology"", ""Dairy Products"", ""Bacteria"", ""Animals"", ""Milk"", ""Dairying"", ""Artificial Intelligence"", ""Machine Learning""]", 42874345,https://pubmed.ncbi.nlm.nih.gov/42874345/,Machine learning-driven prediction of Salmonella enterica virulence for enhanced risk assessment in poultry in Mexico.,"Strain-level heterogeneity in Salmonella enterica pathogenic potential is not routinely incorporated into quantitative microbial risk assessment (QMRA), where homogeneous virulence assumptions are commonly applied. This study evaluated whether supervised machine learning (ML)-based genomic classification could provide a structured approach to integrating strain-level variation into risk modeling. Publicly available S. enterica genomes annotated with clinical metadata were used to train classifiers assigning isolates to clinically derived risk categories. Two genomic feature representations were compared: pangenome-wide gene presence/absence matrices and curated functional gene sets. Ensemble models trained under nested cross-validation achieved strong predictive performance, with pangenome-based models outperforming models restricted to curated functional gene sets. Feature importance analyses indicated that accessory genome variation contributed substantially to class discrimination. The trained classifiers were applied to an independent cohort of poultry-derived isolates to examine out-of-sample prediction patterns. Model-derived classifications were subsequently incorporated into an established QMRA framework representing high-risk chicken consumption scenarios in central Mexico. Integration of virulence categories modified dose-response assumptions and redistributed prevalence across risk strata, producing scenario-dependent differences in estimated probabilities of illness and projected case numbers. The median estimated risk varied according to the genomic stratification approach and scenario. For the ML pangenome-based model, risk estimates ranged from 5.84 × 10-10 to 1.02 × 10-3. The ML model based on curated functional genes produced estimates from 5.7× 10-10 to 1.67 × 10-3. These were compared with a previous 13-gene model (5.72 × 10-10 to 1.70 × 10-3) and a serotype-based approach (1.42 × 10-9 to 6.00 × 10-3). These findings indicate that genome-informed stratification can be operationally embedded within QMRA and, importantly, that the way strains are classified directly influences estimated risk levels. Because predicted risks varied substantially depending on whether isolates were categorized as high-, moderate-, or low-virulence, accurate genomic classification is critical for appropriately capturing strain-level heterogeneity can influence predicted public health outcomes within modeled scenarios.","[""Godínez-Oviedo Angélica"", ""García-Huerta Santiago"", ""Hernández-Iturriaga Montserrat"", ""Gómez-Baltazar Adrián""]",International journal of food microbiology,2026-10-10,10.1016/j.ijfoodmicro.2026.112114,"[""Journal Article""]","[""Animals"", ""Salmonella enterica"", ""Risk Assessment"", ""Mexico"", ""Virulence"", ""Machine Learning"", ""Chickens"", ""Poultry"", ""Salmonella Infections, Animal"", ""Poultry Diseases"", ""Predictive Learning Models"", ""Genome, Bacterial"", ""Classification Algorithms"", ""Prediction Algorithms""]", 42962737,https://pubmed.ncbi.nlm.nih.gov/42962737/,"LILRB3 as an ulcerative colitis-associated molecular biomarker: integrated transcriptomic, single-cell, and experimental validation.","OBJECTIVE: Ulcerative colitis (UC) is a chronic inflammatory bowel disease characterized by mucosal inflammation and immune dysregulation. Although macrophage-mediated inflammation is associated with UC progression, its molecular basis remains incompletely understood. This study integrated bioinformatic analyses and experimental validation to identify UC-associated molecular biomarkers and explore their roles in immune dysregulation. METHODS: Bulk transcriptomic datasets were integrated for differential expression analysis, weighted gene co-expression network analysis (WGCNA), and multi-algorithm machine learning. The machine-learning model was evaluated in independent validation cohorts, and a nomogram was subsequently constructed and evaluated in an additional colonic-tissue cohort. Immune infiltration, single-cell RNA sequencing, cell-cell communication, pseudotime analysis, and scTenifoldKnk-based virtual knockout were used to investigate hub genes. RT-qPCR, WB, and IHC validated the findings in a DSS-induced mouse model and clinical samples, while siRNA-mediated LILRB3 knockdown was performed in THP-1-derived macrophages. RESULTS: Differential expression analysis and WGCNA identified 55 candidate genes. Through screening of 127 machine-learning algorithm combinations, DPP10, LILRB3, MTMR11, and SLC6A14 were identified as UC-associated molecular biomarkers. A separately constructed logistic regression nomogram showed an apparent AUC of 0.989. LILRB3 was predominantly expressed in myeloid cells, particularly M1 macrophages, and was associated with inflammatory immune cell infiltration. Cell-cell communication analysis revealed that LILRB3⁺ M1 macrophages exhibited active intercellular communication involving MHC-II-CD4 and other inflammation-related signaling pathways. Experimental validation showed increased PirB expression in the colonic tissues of DSS-induced colitis mice and elevated LILRB3 expression in human UC colonic tissues. Knockdown of LILRB3 in THP-1-derived macrophages significantly reduced TNF-α, IL-6, and IL-1β mRNA expression. CONCLUSION: LILRB3 is a candidate UC-associated molecular biomarker in colonic tissue and may be associated with proinflammatory macrophage remodeling and myeloid immune dysregulation, warranting mechanistic investigation.","[""Yang Bo"", ""Sun Dongjie"", ""Liu Zhu"", ""Zhou Lili""]",Gene,2026-10-30,10.1016/j.gene.2026.150471,"[""Journal Article""]","[""Colitis, Ulcerative"", ""Humans"", ""Animals"", ""Mice"", ""Biomarkers"", ""Transcriptome"", ""Macrophages"", ""Receptors, Immunologic"", ""Gene Expression Profiling"", ""Machine Learning"", ""Male"", ""Gene Regulatory Networks""]", 42951037,https://pubmed.ncbi.nlm.nih.gov/42951037/,Speech-based differentiation of unipolar and bipolar depression across scripted reading and interview questions.,"BACKGROUND: Differentiating bipolar depression (BPD) from unipolar depression (UPD) remains clinically challenging. We examined whether acoustic and linguistic speech features could support this distinction and whether their informativeness varied across scripted reading and interview questions. METHODS: This study included 109 healthy controls (HCs), 52 patients with UPD, and 53 patients with BPD experiencing depressive episodes. Participants completed a scripted reading task and an interview-based speech task comprising 16 questions. Acoustic features were extracted from all audio recordings, and linguistic features were derived from interview transcripts. Group differences were examined using robust analysis of covariance with global false discovery rate correction. Two-level stacking models with fold-wise sex adjustment of acoustic features evaluated classification performance and importance of speech features and interview-based speech elicitation prompts. RESULTS: Multiple acoustic and linguistic features showed significant group differences in interview responses, whereas no acoustic features differed significantly during scripted reading. The best-performing models achieved F1 scores of 0.6655 for three-class classification (HC, UPD and BPD) and 0.6430 for binary classification (UPD and BPD). Within interview-based speech, statistical and machine learning analyses identified Average Pause Duration, Phonation Ratio, and quantity-related expressions as particularly informative features. Two interview questions about (hypo-)manic experiences contributed most to differentiating BPD from UPD. CONCLUSIONS: Acoustic and linguistic speech features may help differentiate BPD from UPD during depressive episodes, with their informativeness varying across speech tasks and interview questions. Our findings suggest that targeted interview-based speech assessment may have potential as a non-invasive adjunct to conventional clinical evaluation.","[""Zhong Rongrong"", ""Zhang Yuezhou"", ""Wu Xiaohui"", ""Cui Lvchun"", ""Lin Jiansen"", ""Lv Xue"", ""Chen Jiayue"", ""Chen Shen Tse"", ""Yu Yue"", ""Folarin Amos"", ""Dobson Richard"", ""Chen Jun"", ""Fang Yiru""]",Journal of affective disorders,2026-10-28,10.1016/j.jad.2026.122675,"[""Journal Article""]","[""Humans"", ""Bipolar Disorder"", ""Female"", ""Male"", ""Adult"", ""Diagnosis, Differential"", ""Reading"", ""Interview, Psychological"", ""Middle Aged"", ""Speech"", ""Depressive Disorder"", ""Speech Acoustics"", ""Major Depressive Disorder"", ""Machine Learning""]", 42918352,https://pubmed.ncbi.nlm.nih.gov/42918352/,Machine learning-based computerized adaptive testing of PHQ-8 and GAD-7 in psychiatric outpatients.,"BACKGROUND: Scalable, accurate and efficient monitoring tools are urgently needed to support psychiatric assessment in routine care. This study aimed to evaluate a machine learning-based computerized adaptive testing (ML-CAT) model for the PHQ-8 and GAD-7 in psychiatric outpatient settings. METHODS: Data were obtained from a longitudinal cohort study of patients with depression or anxiety diagnoses. ML-CAT models were developed using the M5P algorithm. Model performance was evaluated via 10-fold cross-validation. Concurrent validity was assessed with Pearson's r, R2, and mean absolute error (MAE). Longitudinal change detection was assessed using paired t-tests and Cohen's d. Diagnostic classification agreement (based on the ≥10 cutoff between assessments) was examined using percentage agreement and Cohen's κ. RESULTS: The ML-CAT administered a median of 2 items per instrument (66-71% item reduction) while achieving near-perfect concurrent validity with full-scale scores (r = 0.996-0.997; R2 = 0.986-0.991; MAE < 0.62). Both ML-CAT versions reliably detected symptom reductions at 6 months and 1 year. Classification agreement between ML-CAT and full-scale versions was high at 6 months and 1 year for PHQ-8 (κ = 0.85-0.93) and GAD-7 (κ = 0.91-0.96). CONCLUSIONS: The ML-CAT model reproduced PHQ-8 and GAD-7 scores with high accuracy and detected meaningful clinical change across two follow-up waves. The resulting adaptive questionnaire can be presented as a flowchart and administered both online and offline, providing real-world clinical evidence for scalable ML-CAT-based psychiatric assessment.","[""Abdin Edimansyah"", ""Jeyagurunathan Anitha"", ""Tan Sing Chik"", ""Mok Yee Ming"", ""Verma Swapna"", ""Feng Mengling"", ""Subramaniam Mythily""]",Journal of affective disorders,2026-10-23,10.1016/j.jad.2026.122660,"[""Journal Article""]","[""Humans"", ""Machine Learning"", ""Female"", ""Male"", ""Adult"", ""Outpatients"", ""Longitudinal Studies"", ""Middle Aged"", ""Anxiety Disorders"", ""Reproducibility of Results"", ""Psychometrics"", ""Algorithms"", ""Diagnosis, Computer-Assisted"", ""Adaptive Algorithms"", ""Psychiatric Status Rating Scales""]", 42918350,https://pubmed.ncbi.nlm.nih.gov/42918350/,Circulating immune cells phenotypes discriminate bipolar disorder from major depressive disorder during a depressive episode: A machine learning elastic-net approach.,"MDD and BD are characterized by differences in peripheral immune markers, in frequencies of T helper, T cytotoxic, Natural Killer (NK) and T regulatory cells that may contribute to the development of mood disorders. However, few studies have used immune cells populations to discriminate MDD from BD with inconsistent findings. We applied an elastic-net machine learning pipeline predictive model using 5-fold nested cross-validation based on a wide panel of immune cell populations to discriminate between MDD and BD. The sample included 124 MDD and 89 BD hospitalised for depressive episode, and 40 healthy controls (HC). The predictive model achieved an accuracy of 78%, with 76% sensitivity, 80% specificity. The model performed significantly better than chance (p < 0.001). BD diagnosis was associated with higher frequencies of cells populations producing proinflammatory cytokines (i.e. CD4+IFN-γ, CD4+IL-17, CD4+TNF-α, CD8+TNF-α, NK cells/TNF-α), and regulatory cells (FOXP3), whereas MDD diagnosis was associated with elevated frequencies of early activating memory cells both T helper and cytotoxic (i.e. CD4+CM/CD69+, CD4+EM/CD69+, CD8+CM/CD69+), and cytotoxic effector cells producing apoptotic proteins (CD8+EM/Perf+GrzB+, NK cells/Perf+GrzB+). Excluding patients in lithium only early activated and cytotoxic cells differentiated the two diagnoses. Limitations include possible residual instrument-related effects and cryopreservation effects, lack of generalization to other phases of the illness, lack of external validation. Our data confirm that MDD and BD show different immune signatures and suggest a more pronounced alteration of immune cell populations in MDD especially in cytotoxic CD8+ cells. Future studies should address these differences to develop more effective therapies for mood disorders.","[""Poletti Sara"", ""Bravi Beatrice"", ""Calesella Federico"", ""Aggio Veronica"", ""Vai Benedetta"", ""Finardi Annamaria"", ""Mandelli Alessandra"", ""Lorenzi Cristina"", ""Colombo Cristina"", ""Furlan Roberto"", ""Benedetti Francesco""]",Journal of affective disorders,2026-10-23,10.1016/j.jad.2026.122658,"[""Journal Article""]","[""Humans"", ""Bipolar Disorder"", ""Major Depressive Disorder"", ""Female"", ""Male"", ""Machine Learning"", ""Adult"", ""Diagnosis, Differential"", ""Cytokines"", ""Killer Cells, Natural"", ""Middle Aged"", ""Phenotype"", ""T-Lymphocytes, Regulatory"", ""Case-Control Studies"", ""Predictive Learning Models""]", 42914900,https://pubmed.ncbi.nlm.nih.gov/42914900/,An artifact-resilient multimodal bioelectronic patch for monitoring biochemical and motor states in Parkinson's disease.,"Personalized management of Parkinson's disease (PD) requires continuous tracking of motor fluctuations and medication-associated biochemical dynamics during daily living. Here, we report the Biochemical and Inertial Multimodal Assessment Patch (BIMAP), a fully integrated multimodal bioelectronic patch that combines multiplexed sweat biosensing of levodopa (L-Dopa), glucose (Glu), and ascorbic acid (AA) with triaxial accelerometry for biochemical-kinematic monitoring of PD-related states. A microfluidic sampling interface is incorporated to stabilize sweat transport and maintain sensor-fluid contact during motion. We further develop a machine-learning-assisted Multimodal Adaptive Gating Network (MAG-Net) to adaptively weight biochemical and inertial features according to the instantaneous motion state. This strategy reduces the contribution of motion-perturbed electrochemical features, while retaining L-Dopa-related biochemical context for distinguishing medication-associated motor changes from frequency-matched daily activities. On-body feasibility tests involving healthy volunteers and participants with PD showed that the patch could track intake-associated biochemical changes and temporally aligned changes in estimated sweat L-Dopa and motor activity after medication administration. The multimodal signals further enabled exploratory discrimination of four protocol-defined medication-motion states under the tested conditions. These results support the feasibility of artifact-resilient biochemical-kinematic monitoring in wearable settings.","[""Geng Haoyang"", ""Li Shiyao"", ""Liang Changcheng"", ""Min Shunhua"", ""Sun Xin"", ""Peng Yuyan"", ""Ding Shichao"", ""Wang Zhijun"", ""Liu Yibiao"", ""Liu Qingzhou"", ""Xu Tailin""]",Biosensors & bioelectronics,2026-10-22,10.1016/j.bios.2026.119363,"[""Journal Article""]","[""Parkinson Disease"", ""Humans"", ""Biosensing Techniques"", ""Levodopa"", ""Sweat"", ""Ascorbic Acid"", ""Glucose"", ""Wearable Electronic Devices"", ""Accelerometry"", ""Artifacts"", ""Male"", ""Female""]", 42906709,https://pubmed.ncbi.nlm.nih.gov/42906709/,EEG-VTSFN: An enhanced approach for differentiating between unipolar and bipolar mood disorders.,"Accurate differentiation between Bipolar Disorder (BD) and Unipolar Depression (UD) is a significant clinical challenge due to overlapping manifestations and a lack of objective biomarkers. While EEG offers high temporal resolution, its psychiatric utility for diagnostic differentiation remains limited. To address this, we propose the Variational Autoencoder (VAE)-based Time-Space Fusion Network (VTSFN). VTSFN transforms complex EEG signals into latent-space representations to mitigate noise and construct stable latent co-activation patterns. By integrating Long Short-Term Memory (LSTM) and attention mechanisms, the model captures temporal dynamics to identify discriminative patterns. Our results demonstrate that VTSFN achieved the highest accuracy and AUC point estimates among the evaluated models, with a test-set accuracy of 79.3% (95% CI, 65.5%-93.1%) and an AUC of 82.7% (95% CI, 63.2%-98.5%), without significant paired AUC differences at p < 0.01. Validation in a diagnostic-switch (tBD) cohort (n = 22) identified 17/22 converters (sensitivity, 77.3%; 95% CI, 59.1%-90.9%), requiring further longitudinal validation. Exploratory feature analysis revealed that, compared with UD patients, BD and tBD patients exhibited weaker latent co-activation in right parieto-temporal and temporo-occipital regions, alongside stronger latent co-activation in left frontal-right parietal and temporal regions. These model-derived channel similarity patterns require independent validation.","[""Zhao Xixi"", ""Cui Yi"", ""Zhao Tong"", ""Wang Bin"", ""Wang Yingtan"", ""Chen Fangyu"", ""Zhang Haoran"", ""Cui Yifan"", ""Liu Yuci"", ""Zhang Yu"", ""Sha Sha"", ""Zhang Ling"", ""Hong Bo"", ""Lu Bai"", ""Yan Yuxiang"", ""Ren Yanping"", ""Wang Gang""]",Journal of affective disorders,2026-10-21,10.1016/j.jad.2026.122655,"[""Journal Article""]","[""Humans"", ""Bipolar Disorder"", ""Electroencephalography"", ""Diagnosis, Differential"", ""Female"", ""Depressive Disorder"", ""Adult"", ""Male"", ""Autoencoder"", ""Long Short Term Memory""]", 42906310,https://pubmed.ncbi.nlm.nih.gov/42906310/,SF2O-SKDCN: Selective kernel attention enabled deep learning framework for cardiovascular disease detection using photoplethysmography signals.,"Cardiovascular Disease (CVD) is a lethal disease that needs early intervention to prevent further harm and eventually safeguard patients' lives. Therefore, CVD detection is critical for effective disease management. Even though practical findings are made using traditional detection methods, they mostly struggle with overfitting risks, generalizability issues, and computational overhead challenges during the detection process. Therefore, the research proposes Sail Tuna Fish Foraging Optimized-Selective Kernel attention-enabled Deep Convolutional Neural Network (SF2O-SKDCN) that utilizes photoplethysmography (PPG) signals, which are extensively used in clinical applications as a screening tool, and the heart rate feature descriptor is introduced to extract the crucial feature from the segmented signal. Further, the Selective Kernel Shuffle (SKS) attention enables the model to focus only on the predominant CVD features, thereby enhancing the interpretability characteristics. In addition, actively tuning the model hyperparameters using the Sail Tuna Fish Foraging Optimization (SF2O) algorithm prevents local convergence and vanishing gradient issues. Experimenting with the MIMIC II dataset, the model yields better results by achieving 98.54% accuracy, 99.07% recall, 98.05% Positive predictive value, and 99.05% Negative predictive value for 90% training.","[""Chavan Pratima M"", ""Joshi Kalyani R"", ""Kamathe Rupali S""]",Biosensors & bioelectronics,2026-10-09,10.1016/j.bios.2026.119264,"[""Journal Article""]","[""Photoplethysmography"", ""Deep Learning"", ""Humans"", ""Animals"", ""Cardiovascular Diseases"", ""Tuna"", ""Signal Processing, Computer-Assisted"", ""Heart Rate"", ""Biosensing Techniques"", ""Convolutional Neural Networks"", ""Algorithms""]", 42901112,https://pubmed.ncbi.nlm.nih.gov/42901112/,Machine learning-empowered nanobody-based dual-mode immunosensor using gold-silver nanoclusters for the accurate detection of Shigella flexneri in food.,"Sensitive and rapid detection of Shigella flexneri (S. flexneri) is of great significance for food safety and public health. In this study, a nanobody-based colorimetric/fluorescent dual-mode immunosensor was developed for the first time, and a machine learning model was integrated for intelligent fusion of the dual signals. A phage-displayed nanobody library was constructed by immunizing Bactrian camels with formalin-inactivated S. flexneri and specific nanobodies were obtained after four rounds of solid-phase panning. Using bovine serum albumin-protected gold-silver bimetallic nanoclusters (Au-Ag NCs) as fluorescent probes, a dual-signal output platform was established. Under optimized conditions, the colorimetric mode achieved a limit of detection (LOD) of 2.992 × 104 CFU/mL, and exhibited no cross-reactivity with other common foodborne pathogens. The LOD in the fluorescent mode was as low as 4.074 × 103 CFU/mL, which is 7.3 times lower than that of the colorimetric mode. A machine learning model combining K-Nearest Neighbors classification and Random Forest regression was employed for intelligent calibration of the dual signals, significantly improving the reliability and stability of the assay. With the assistance of the machine learning model, the coefficient of determination (R2) between predicted and actual concentrations reached 0.9953. Spike-and-recovery experiments in food samples yielded recovery rates ranging from 81.53% to 109.46%, demonstrating high accuracy. This study provides a novel, rapid, accurate, and environmentally friendly strategy for the detection of S. flexneri, holding great promise for applications in food safety monitoring.","[""Fan Xueyi"", ""Zheng Kai"", ""Wang Mengnan"", ""Ge Guangsheng"", ""Miao Yuxin"", ""Yao Qing"", ""Chen Pengyu"", ""Ma Min"", ""Wang Jianlong"", ""Wang Yanru""]",Biosensors & bioelectronics,2026-10-20,10.1016/j.bios.2026.119352,"[""Journal Article""]","[""Shigella flexneri"", ""Biosensing Techniques"", ""Machine Learning"", ""Silver"", ""Gold"", ""Metal Nanoparticles"", ""Animals"", ""Single-Domain Antibodies"", ""Limit of Detection"", ""Immunoassay"", ""Camelus"", ""Colorimetry"", ""Food Microbiology"", ""Dysentery, Bacillary""]", 42892230,https://pubmed.ncbi.nlm.nih.gov/42892230/,"Flexible and stretchable bioelectronics for sports wearable monitoring: Materials, Device Engineering, and Intelligent Wearable Platforms.","Modern sports science is undergoing a paradigm shift from subjective coaching to objective, data-driven precision training. Consequently, wearable electronics for acquiring precise data during sports activities are attracting attention. However, conventional rigid wearables inherently suffer from mechanical mismatches between rigid wearables and the soft, deformable human skin, inducing severe motion artifacts and restricting natural biomechanics during high-intensity athletic activities. To address these limitations from a comprehensive perspective, this review systematically summarizes and analyzes recent advancements in next-generation flexible and stretchable bioelectronics tailored specifically for sports applications. We systematically examine the substrate materials, structural engineering designs, and device-level optimization strategies required to withstand repetitive strain, shear, and sweat-intensive environments. Building upon these conformable bio-interfaces, we discuss three core sensing modalities: electrochemical sensors for real-time analysis of sweat biomarkers, biophysical sensors for continuous electrophysiological and hemodynamic monitoring, and electromechanical sensors for high-fidelity kinematic and impact tracking. Furthermore, we discuss body-derived energy harvesting strategies to power sports wearables and wireless communication systems for continuous athletic monitoring. The importance of artificial intelligence and machine learning is increasing, as these techniques enable the removal of motion artifacts and facilitate objective analysis of athletic data for training and health monitoring applications. By evaluating targeted applications across aerobic and anaerobic exercise, cognitive state assessment, and post-exercise recovery, this review demonstrates how next-generation sports wearables contribute to the shift in sports science from reactive treatment toward proactive, data-driven performance enhancement and injury prevention.","[""Lee Junseok"", ""Lee Yena"", ""Shin Jongwoon"", ""Lee Doohyun J"", ""Yoon Hyosang"", ""Lee Ju Young"", ""Yu Ki Jun"", ""Sang Mingyu""]",Biosensors & bioelectronics,2026-10-09,10.1016/j.bios.2026.119323,"[""Journal Article"", ""Review""]","[""Humans"", ""Wearable Electronic Devices"", ""Biosensing Techniques"", ""Equipment Design"", ""Sports"", ""Monitoring, Physiologic"", ""Sweat"", ""Digital Health""]", 42887283,https://pubmed.ncbi.nlm.nih.gov/42887283/,Site-regulated CoNFs-NH2@PtNPs nanozyme-based all-in-one portable detection of organophosphorus pesticides integrating deep learning and hardware logic gates.,"Widespread use of organophosphorus pesticides (OPs) has led to pervasive environmental residues and potential human health hazards, making rapid, sensitive, and portable on-site OPs detection urgently needed. In this work, an amino-functionalized cobalt nanoflower-supported platinum nanoparticle composite (CoNFs-NH2@PtNPs) was successfully fabricated. Combined with steady-state kinetic experiments and density functional theory calculations, the modulation mechanism of amino modification on the electron-donating ability of Pt active sites and oxygen activation performance was systematically clarified. Taking advantage of the inherent oxidase-like activity and fluorescence properties of CoNFs-NH2@PtNPs, a colorimetric-fluorescent dual-mode sensor was constructed based on acetylcholinesterase inhibition for high-efficiency OPs detection. Furthermore, a low-cost integrated portable detection platform was developed by integrating 3D-printed dark-box hardware, MCU control modules, smartphone imaging technology, and self-built sensing algorithms. This platform realizes intuitive qualitative grading via LED indicator lights based on solution RGB signal changes and intelligent quantitative detection through a customized smartphone mini-program. The established all-in-one detection system effectively addresses the poor portability and inferior real-time performance of traditional OPs detection methods, providing a reliable and practical strategy for field pesticide residue monitoring.","[""Zhang Yining"", ""Gao Di"", ""He Panrui"", ""Peng Huajie"", ""Xu Peng"", ""Qiu Ping""]",Biosensors & bioelectronics,2026-10-13,10.1016/j.bios.2026.119335,"[""Journal Article""]","[""Biosensing Techniques"", ""Organophosphorus Compounds"", ""Pesticides"", ""Colorimetry"", ""Deep Learning"", ""Platinum"", ""Smartphone"", ""Metal Nanoparticles"", ""Humans"", ""Limit of Detection"", ""Equipment Design""]", 42877889,https://pubmed.ncbi.nlm.nih.gov/42877889/,DiFVM: A Vectorized Graph-Based Finite Volume Solver for Differentiable CFD on Unstructured Meshes.,"Differentiable programming has emerged as an important tool for gradient-based inverse problems and end-to-end hybrid physics-machine learning in computational fluid dynamics. Existing differentiable CFD platforms predominantly rely on structured Cartesian grids, limiting their direct application to geometrically complex domains where body-conforming unstructured discretizations are essential. We present DiFVM, the first GPU-native, end-to-end differentiable finite-volume (FV) CFD solver for incompressible flow on unstructured polyhedral meshes. The central methodological contribution is a graph-based computational reformulation of the complete unstructured FV/PISO solution procedure: by expressing FV operators as static scatter/gather primitives on the mesh connectivity graph, DiFVM transforms irregular unstructured connectivity into a GPU-compatible computational representation. All operations are implemented in JAX/XLA, providing just-in-time compilation, operator fusion, and automatic differentiation through the complete simulation pipeline. Differentiable Windkessel outlet boundary conditions are provided for cardiovascular applications, and DiFVM accepts standard OpenFOAM case directories without modification for seamless adoption in existing workflows. Forward accuracy is assessed through analytical and canonical benchmark problems, comparisons with published reference data, and independent solver comparisons, while end-to-end differentiability is demonstrated through proof-of-concept inverse modeling examples. DiFVM provides a differentiable CFD framework for gradient-based optimization, inverse modeling and future physics-integrated machine learning on complex unstructured geometries.","[""Du Pan"", ""Li Yongqi"", ""Xu Mingqi"", ""Wang Jian-Xun""]",Journal of computational physics,2026-09-25,10.1016/j.jcp.2026.115429,"[""Journal Article""]",,PMC13668260 42863795,https://pubmed.ncbi.nlm.nih.gov/42863795/,Machine learning in the prediction of binge drinking: A systematic review.,"BACKGROUND: Binge drinking (BD) refers to episodes of excessive alcohol consumption, which are associated with various acute and long-term consequences. Despite its high prevalence and negative impacts, predictive features for BD remain lacking. This study aims to review machine learning (ML) research on BD, comparing different algorithms and types of data, trying to identify promising models and variables that may help identify individuals at risk. METHODS: A systematic search was performed in PubMed, Embase, and Scopus to identify relevant studies published before July 2, 2026. 21 studies were included in data extraction process. RESULTS: The results suggest that impulsive factors, substance-use variables, emotional distress, as well as demographic features, particularly gender and age, play a significant role in predicting BD. Moreover, emerging methods that use real-time data recorded via mobile apps showed promise. Conversely, findings from neural data yielded mixed results, with no single brain area consistently associated with BD. The qualitative comparison did not show any specific ML algorithm consistently outperforming the others across studies; however, the predictive performance of ML approaches tended to improve when different data types were combined. LIMITATIONS: The analysis of the available literature was restricted due to the non-homogeneous characteristics of studies, mainly in terms of methodology, which did not allow us to draw comprehensive conclusions. CONCLUSIONS: The comparison of available ML studies showed how promising risk factors can be combined within multivariate models to increase their predictive value. Future large-scale studies using multisource data may better define risk profiles to foster prevention strategies.","[""Colli Chiara"", ""Pigoni Alessandro"", ""Cellini Paolo"", ""Peyvandi Aida"", ""Enrico Gabriele"", ""Moltrasio Chiara"", ""Brambilla Paolo"", ""Delvecchio Giuseppe""]",Journal of affective disorders,2026-10-12,10.1016/j.jad.2026.122617,"[""Journal Article"", ""Systematic Review""]","[""Binge Drinking"", ""Machine Learning"", ""Humans"", ""Predictive Learning Models"", ""Prediction Algorithms"", ""Classification Algorithms""]", 42863404,https://pubmed.ncbi.nlm.nih.gov/42863404/,Cultivar differentiation of fritillary using machine learning model-driven untargeted metabolomics.,"Fritillaria cirrhosa (F. cirrhosa), a medicinal and edible herbs, is highly regarded for its remarkable efficacy in relieving cough and eliminating phlegm. However, surging demand and wild resource shortages have caused widespread adulteration of F. thunbergii with F. ussuriensis. Therefore, conducting a comparative study of these types of fritillaries may be helpful for the quality control and identification of the F. cirrhosa medicinal material. This study aims to conduct a comparative study of the chemical components of three types of fritillaries by integrating plant metabolomics combined with machine learning, from the perspectives of component analysis, differential identification, and quantitative analysis.The chemical components of fritillaries were comprehensively characterized using UHPLC-Q-Exactive Orbitrap MS technology, and a total of 220 chemical components were identified. Combined with chemometrics methods,untargeted metabolomics and six machine learning models, eight differential components were further selected, and their contents were determined. Among them, sipeimine, edpetiline, solanidine, ebeiedinone, and peimisine may serve as differential components for distinguishing F. cirrhosa from F. thunbergii and F. ussuriensis, providing chemical basis for the authenticity identification and quality control of F. cirrhosa.","[""Xu Miao"", ""Wang Xiaoxuan"", ""Kang Mingqin"", ""Zhang Lijuan"", ""Li Guohui"", ""Xu Yanyan"", ""Song Lili""]",Journal of pharmaceutical and biomedical analysis,2026-10-08,10.1016/j.jpba.2026.117757,"[""Journal Article""]","[""Metabolomics"", ""Fritillaria"", ""Machine Learning"", ""Chromatography, High Pressure Liquid"", ""Quality Control"", ""Drugs, Chinese Herbal"", ""Plants, Medicinal""]", 42858627,https://pubmed.ncbi.nlm.nih.gov/42858627/,An Ago-mediated duplex biosensing platform with orthogonal electrochemical coding for microRNA detection based on TDN interfaces and UiO-66-NH2 nanocarriers.,"Circulating microRNAs (miRNAs) are colorectal cancer (CRC) biomarkers, but their short length, high sequence homology, and low abundance make sensitive and multiplexed detection difficult. Electrochemical biosensing offers a route for miRNA analysis, yet duplex detection remains limited by the need for reporter pairs with well-separated signals and minimal crosstalk. Here, we report an Ago-mediated Duplex Orthogonal Readout for microRNA Electrochemical detection (ADORE) for simultaneous detection of CRC-related miRNAs, miR-21 and miR-92a. ADORE uses methylene blue and L-012 as an orthogonal electrochemical code pair, validating L-012, a luminol derivative used in chemiluminescence and electrochemiluminescence, as an electrochemical reporter. The sensing interface integrates tetrahedral DNA nanostructures with reporter-loaded UiO-66-NH2 carriers to stabilize probe organization and amplify electrochemical responses. Target miRNAs first initiate an exponential amplification reaction (EXPAR) to generate guide DNAs. These guides then program Clostridium butyricum Argonaute (CbAgo) for sequence-specific cleavage at the interface, producing target-dependent signal-off responses. ADORE achieves limits of detection of 1.85 aM for miR-21 and 0.51 aM for miR-92a. To improve analysis of serum samples, multidimensional differential pulse voltammetry-derived features were integrated into a support vector machine model, increasing diagnostic performance from an area under the receiver operating characteristic curve (AUC) of 0.88 using single-feature analysis to 0.94 using multifeature classification in serum discrimination. This work provides an integrated strategy combining Ago-mediated signal transduction, orthogonal electrochemical coding, nanostructured interface engineering, and machine learning for duplex serum miRNA analysis and CRC-related biomarker detection, supporting development of multiplexed electrochemical liquid-biopsy assays for CRC screening.","[""Li Hanxue"", ""Wu Kexin"", ""Li Hanhua"", ""Kong Jilie"", ""Chen Hui""]",Biosensors & bioelectronics,2026-10-08,10.1016/j.bios.2026.119309,"[""Journal Article""]","[""Biosensing Techniques"", ""Electrochemical Techniques"", ""Humans"", ""MicroRNAs"", ""DNA Nanostructures"", ""Colorectal Neoplasms"", ""Limit of Detection"", ""Argonaute Proteins"", ""Metal-Organic Frameworks"", ""Phthalic Acids""]", 42843470,https://pubmed.ncbi.nlm.nih.gov/42843470/,A two-stage diagnostic classification framework integrating heart rate variability and clinical features to identify recent suicide-related status in psychiatric inpatients.,"BACKGROUND: Accurate identification of recent suicide-related risk remains challenging in psychiatric settings. Although clinical assessment is essential, objective physiological indicators such as heart rate variability (HRV) may provide complementary information. This study developed a two-stage machine-learning framework integrating clinical features and HRV to identify recent suicide-related status among psychiatric inpatients. METHODS: This retrospective study included 619 psychiatric inpatients who underwent standardized clinical assessment and 5-min resting HRV recording. Patients were classified as no suicidal ideation or behavior (no SI/SB; n = 416), suicidal ideation without suicidal behavior (SI; n = 137), or suicidal behavior (SB; n = 66) based on suicide-related information within the month before admission. Stage 1 classified patients with any recent SI/SB versus no SI/SB, and Stage 2 further classified recent SB versus SI among patients with suicide-related risk. Demographic-clinical, demographic-HRV, and combined feature sets were evaluated using multiple machine learning algorithms with repeated nested cross-validation. RESULTS: The combined feature set achieved the best overall performance. In Stage 1, the best combined model showed good discrimination (AUC = 0.894; AUPRC = 0.831). In Stage 2, the best combined model showed moderate performance (AUC = 0.778; AUPRC = 0.686). Depressive symptom severity, self-harm, past suicide attempt, younger age, diagnosis, and reduced vagally mediated HRV contributed to identifying any SI/SB, whereas past suicide attempt was the leading predictor of SB. CONCLUSIONS: Integrating clinical and HRV features may support the identification and stratification of recent suicide-related status in psychiatric inpatients. HRV provided modest complementary information, particularly for broad SI/SB identification, but external prospective validation is required.","[""Wang Yi"", ""Shi Shanshan"", ""Lu Yang"", ""Yu Hao"", ""Zhong Baoyi"", ""Gao Lanyang"", ""Xiang Bo"", ""Wang Tingting"", ""Liu Kezhi""]",Journal of affective disorders,2026-10-07,10.1016/j.jad.2026.122594,"[""Journal Article""]","[""Humans"", ""Female"", ""Heart Rate"", ""Adult"", ""Male"", ""Suicidal Ideation"", ""Machine Learning"", ""Inpatients"", ""Retrospective Studies"", ""Mental Disorders"", ""Middle Aged"", ""Suicide, Attempted"", ""Suicide"", ""Classification Algorithms""]", 42843106,https://pubmed.ncbi.nlm.nih.gov/42843106/,Efficient capture and proteomic profiling of extracellular vesicles with magnetic cilia channel chip for pre- and postoperative assessment of prostate cancer.,"Extracellular vesicles (EVs) are promising non-invasive biomarkers, but their clinical application is limited by inefficient isolation and downstream analysis. Here, we developed a biomimetic magnetic cilia microfluidic chip (MagCilia-chip) for EV capture and proteomic profiling from urine. Tim4-functionalized magnetic microcilia specifically captured phosphatidylserine-positive EVs and underwent magnetic-field-driven three-dimensional rotation to enhance EV-microstructure interactions. Compared with ultracentrifugation, the MagCilia-chip operated in 3D rotation mode achieved a recovery rate of 85% versus 27%, with 5.32-fold higher EV yield and 3.13-fold higher purity. On-chip lysis and digestion followed by mass-spectrometry analysis identified 2903 EV-derived proteins from 50 μL of urine. In the discovery cohort of 12 prostate cancer (PCa) and 8 post-treatment (PT) samples, 368 differentially expressed EV proteins were identified, yielding a five-protein candidate panel. Independent parallel reaction monitoring analysis of 19 PCa and 9 PT samples confirmed differential expression of C9, COL6A3, and VTN. A three-protein panel achieved area-under-the-curve values of 0.78-0.95 across ten machine-learning classifiers using leave-one-out cross-validation. This platform provides an efficient strategy for urinary EV isolation and proteomic analysis and supports further evaluation of EV-derived candidate biomarkers for PCa treatment.","[""Song Xiaorong"", ""Li Mingyang"", ""Sun Jie"", ""Li Xiang"", ""Deng Yuanyuan"", ""Chang Wenya"", ""Cai Yuhan"", ""Li Pengli"", ""Gu Zhongze"", ""Wang Shangqian"", ""Xie Zhuoying""]",Biosensors & bioelectronics,2026-09-20,10.1016/j.bios.2026.119252,"[""Journal Article""]","[""Humans"", ""Male"", ""Prostatic Neoplasms"", ""Extracellular Vesicles"", ""Proteomics"", ""Biomarkers, Tumor"", ""Biosensing Techniques"", ""Lab-On-A-Chip Devices"", ""Equipment Design"", ""Proteome""]",