Archive 4 checkpoints (1-4)

.gitattributes

@@ -268,3 +268,4 @@ paper-releases/pbisc-diffusion/updates/20261015T100456Z-figure1-diffusion-readab
 paper-releases/pbisc-diffusion/updates/20261015T100456Z-figure1-diffusion-readable-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261015_figure1_focused.pdf filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261015T100456Z-figure1-diffusion-readable-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261015_figure1_focused.pptx filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261015T100456Z-figure1-diffusion-readable-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/figures/figure1_focus_20261015/overview/experimental_overview.pdf filter=lfs diff=lfs merge=lfs -text
+paper-releases/pbisc-diffusion/updates/20261015T121654Z-x1-nine-core40-deck/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/x1_priority_20261015/builds/20261015T120743086141Z/NEW_RESULTS_REVIEW.pdf filter=lfs diff=lfs merge=lfs -text

paper-releases/pbisc-diffusion/updates/20261015T121654Z-x1-nine-core40-deck/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/x1_priority_20261015/builds/20261015T120743086141Z/INDIVIDUAL_DONOR_INTERVALS.csv

@@ -0,0 +1,29 @@
+unit,cohort,arm,metric,status,estimate,ci_low,ci_high,n_donors,n_boot,bootstrap_unit
+giroux2022-disease-draw0,giroux2022,lora,comp_l1,ok,0.2718126482768173,0.2360751663474706,0.307550130206164,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,comp_l1_obs_loo_null,ok,0.3634285641471839,0.2823277246132213,0.4744704097690243,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,coverage,ok,0.9616666666666666,0.9183333333333332,0.99,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,coverage_depth_matched,ok,0.9655555555555556,0.925,0.9988888888888888,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_B cell,ok,0.044361111111111,0.0244999999999999,0.0642222222222221,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_DC,ok,0.0229999999999999,0.0159444444444444,0.0296666666666666,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_Monocyte,ok,0.1935555555555555,0.1177777777777777,0.2850555555555555,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_NK cell,ok,0.0650277777777777,0.035611111111111,0.0961944444444444,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_Other,ok,0.0331388888888888,0.0277499999999999,0.0370555555555555,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,gen_T cell,ok,0.6409166666666667,0.52625,0.7107222222222223,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,median_umi_gen,ok,472.5833333333333,415.1666666666666,530.0,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,median_umi_obs,ok,505.0625,447.33333333333337,562.7916666666667,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,mono_pct_gen,ok,19.355555555555554,11.77777777777778,28.505555555555556,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,mono_pct_obs,ok,21.859692610326228,13.243292142721568,30.476093077930884,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,n_generated_cells,ok,1500.0,1500.0,1500.0,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,n_observed_cells,ok,9328.375,7652.375,10654.166666666668,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,n_reference_cells,ok,150.0,150.0,150.0,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_B cell,ok,0.0753064588521087,0.0419286709835125,0.1086842467207049,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_DC,ok,0.0153516390435263,0.0106309366765227,0.0192269645792149,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_Monocyte,ok,0.2185969261032622,0.1324329214272156,0.3047609307793089,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_NK cell,ok,0.1095351653645975,0.0774535205955461,0.1416168101336488,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_Other,ok,0.0533811320440611,0.0234015210252346,0.0750197665225483,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,obs_T cell,ok,0.5278286785924438,0.4290086021069872,0.5932830183965533,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,pb_own_rank,ok,2.75,1.4166666666666663,4.333333333333334,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,pb_r,ok,0.863823778310286,0.8555672662563876,0.8720802903641844,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,pb_top1,ok,0.4166666666666666,0.0833333333333333,0.8333333333333333,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,precision,ok,1.0,1.0,1.0,4,2000,unique_donor_all_visits
+giroux2022-disease-draw0,giroux2022,lora,precision_depth_matched,ok,0.9947222222222222,0.9913888888888888,0.9972222222222222,4,2000,unique_donor_all_visits

paper-releases/pbisc-diffusion/updates/20261015T121654Z-x1-nine-core40-deck/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/x1_priority_20261015/builds/20261015T120743086141Z/NEW_RESULTS_REVIEW.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7b0398a4c5f41fb0335eca56bea2f448f5faa6aa10d5e539a1a967da7fa8b28
+size 22980906

paper-releases/pbisc-diffusion/updates/20261015T121654Z-x1-nine-core40-deck/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/x1_priority_20261015/builds/20261015T120743086141Z/page_map.csv

@@ -0,0 +1,170 @@
+page,parent_156_page,group,title,addition
+1,1.0,Introduction,PBISC-Diffusion: manuscript evidence,
+2,2.0,Introduction,Figures 1-6: the expanded evidence map,
+3,3.0,Introduction,How this expanded deck is organized,guide:scope
+4,4.0,Introduction,Manuscript map: the six figures,
+5,5.0,Figure 1,Figure 1. Overview and corpus,
+6,6.0,Figure 1,"Figure 1: overview, corpus and denoising",
+7,7.0,Figure 1,Experimental overview,
+8,8.0,Figure 1,Pretraining data: 33.14 million training cells,figure1b:focused_training_corpus
+9,9.0,Figure 1,Training corpus: complete counts and minor categories,figure1b:training_corpus
+10,10.0,Figure 1,"Full source corpus: training, validation and test",strengthening:corpus:0
+11,11.0,Figure 1,From 41.40M source cells to 33.14M training cells,strengthening:corpus:1
+12,12.0,Figure 1,Recipe K and training exposure are different counts,strengthening:corpus:2
+13,13.0,Figure 1,Corpus scale and generation workflow,
+14,14.0,Figure 1,From Gaussian noise to generated cell states,enrichment:trajectory_four_pca_frames
+15,15.0,Figure 2,Figure 2. SLE model ablations,
+16,16.0,Figure 2,Figure 2: architecture and gene context,
+17,17.0,Figure 2,Where context width and trunk width change,enrichment:arch_context_vs_trunk
+18,18.0,Figure 2,Predicted and oracle hints: one trained denoiser,enrichment:arch_hint_injection
+19,19.0,Figure 2,Architecture: full and late validation progression,
+20,20.0,Figure 2,Width & context ablation: benchmark,
+21,21.0,Figure 2,Architecture arms: dimensions and endpoints,enrichment:arch_width_arm_register
+22,22.0,Figure 2,Gene context: 2k / 3k / 6k / 15k,
+23,23.0,Figure 2,Gene panels: progression relative to each endpoint,enrichment:arch_panel_loss_progression
+24,24.0,Figure 2,Gene-panel arms: exact sizes and training provenance,enrichment:arch_panel_arm_register
+25,25.0,Figure 2,All ablation arms: parameters and training time,enrichment:arch_all_arm_compute
+26,26.0,Figure 2,Real-cell repeat: a sampling reference,enrichment:real_cell_repeat
+27,27.0,Figure 2,Native-pseudobulk model comparisons,review:6
+28,28.0,Figure 3,Figure 3. Real-bulk adaptation,
+29,29.0,Figure 3,Figure 3: the paired-adaptation story,
+30,30.0,Figure 3,Bulk and pooled cells: different expression values,
+31,31.0,Figure 3,Paired calibration reduces same-donor expression error,enrichment:mismatch_same_donor_errors
+32,32.0,Figure 3,Consequence: cells off-manifold,
+33,33.0,Figure 3,Adaptation recovers the observed cell mixture,strengthening:healthy:0
+34,34.0,Figure 3,Adaptation: mixture error and expression agreement,strengthening:healthy:1
+35,35.0,Figure 3,Healthy paired data: full FT and LoRA,guide:availability
+36,36.0,Figure 3,Full FT and LoRA ranks: quality and parameters,review:8
+37,37.0,Figure 3,What does “all pairs” actually mean?,strengthening:healthy:2
+38,38.0,Figure 3,Healthy pair budget: correlation in all cell types,all_celltypes:correlation
+39,39.0,Figure 3,Healthy pair budget: error in all cell types,all_celltypes:mae
+40,40.0,Figure 3,"Five pairs, no extra validation: LoRA vs scSemi",
+41,41.0,Figure 3,Composition: error and donor-level wins,review:13
+42,,Figure 3,Strict paired budget: held-out reconstruction,x1_priority:healthy_budget
+43,,Figure 3,Strict support: composition bias across all six types,x1_priority:healthy_types
+44,42.0,Figure 4,Figure 4. Disease signal preservation,
+45,43.0,Figure 4,Figure 4: disease signal preservation,
+46,44.0,Figure 4,Disease signals with five training donors,
+47,45.0,Figure 4,5-fold CV: disease signal,
+48,46.0,Figure 4,Disease genes at every list length (CAT),
+49,47.0,Figure 4,"Cell-type deconvolution, same pairs: disease signal",
+50,48.0,Figure 4,Disease gene rankings: complete CV comparison,strengthening:disease_metrics:0
+51,49.0,Figure 4,Disease integrity depends on cohort and endpoint,strengthening:disease_metrics:1
+52,50.0,Figure 4,Five-pair Sepsis: metrics across all support draws,strengthening:disease_metrics:2
+53,,Figure 4,Sepsis: strict five-pair disease integrity,x1_priority:disease_ye2026_sepsis
+54,,Figure 4,Imperial ESKD: strict five-pair disease integrity,x1_priority:disease_imperial_eskd
+55,,Figure 4,Giroux: strict five-pair disease integrity,x1_priority:disease_giroux2022
+56,,Figure 4,CHILI: completed core disease metrics,x1_priority:disease_ji2025_chili
+57,51.0,Figure 5,Figure 5. Rare-population identity,
+58,52.0,Figure 5,Figure 5: paired rare-population identity,
+59,53.0,Figure 5,Figure 5 targets begin with the source paper,strengthening:rare_literature:0
+60,54.0,Figure 5,Ye sepsis: published markers in paired scRNA,strengthening:rare_literature:1
+61,55.0,Figure 5,Transcriptional programs and rare-population identity,strengthening:rare_literature:2
+62,56.0,Figure 6,Figure 6. Published-finding reproduction,
+63,57.0,Figure 6,Figure 6: published-finding reproduction,
+64,58.0,Figure 6,Figure 6: what the reproduction must establish,guide:figure6
+65,59.0,Supplementary,Supplementary evidence,
+66,60.0,Supplementary,Supplementary evidence map,
+67,61.0,Supp S01,Corpus reference: assay and source composition,
+68,62.0,Supp S01,Backbone: SDD → PBISC-Diffusion,
+69,63.0,Supp S01,Bulk conditioning,
+70,64.0,Supp S01,Denoiser: joint value + support diffusion,
+71,65.0,Supp S01,Fine-tuning with paired bulk + sc,
+72,66.0,Supp S01,Evaluation design,
+73,67.0,Supp S01,How we score generated cells,
+74,68.0,Supp S02,Does the model follow its bulk conditioning?,healthy_sle_conditioning_interventions
+75,69.0,Supp S02,SLE benchmark: PBISC vs scSemiProfiler,
+76,70.0,Supp S02,SLE benchmark: per cell type,
+77,71.0,Supp S02,Disjoint-cell controls test the bulk-broadcast baseline,healthy_sle_disjoint_truth_control
+78,72.0,Supp S02,SLE: what is added beyond the bulk broadcast?,disease_sle_broadcast
+79,73.0,Supp S03,Same shift in three paired cohorts,
+80,74.0,Supp S03,Input corrections on one real-reference UMAP,healthy_stored_real_reference_umap
+81,75.0,Supp S03,Where generated-cell density differs from real cells,healthy_saved_density_difference
+82,76.0,Supp S03,Input calibration changes cell-type composition,healthy_input_control_composition
+83,77.0,Supp S03,"Same noise, different input: matched-cell shifts",healthy_stored_matched_trajectories
+84,78.0,Supp S03,Matched noise separates cell state from cell identity,healthy_matched_similarity
+85,79.0,Supp S03,Input changes can switch the generated cell label,healthy_label_transitions
+86,80.0,Supp S03,Nearest real cells: distance and label agreement,healthy_nearest_real_by_type
+87,81.0,Supp S03,Composition agreement does not ensure state agreement,healthy_within_type_centroids
+88,82.0,Supp S03,Gene-length effects in the bulk input,healthy_gene_length_input
+89,83.0,Supp S03,Gene-length effects within label-stable cells,healthy_gene_length_stable
+90,84.0,Supp S03,Raw and TPM shifts persist within cell labels,healthy_raw_tpm_top_genes
+91,85.0,Supp S03,Reference calibration leaves specific gene-state shifts,healthy_refcal_top_genes
+92,86.0,Supp S03,Donor trait axes and donor identification differ,healthy_donor_axes
+93,87.0,Supp S03,Reference choice changes different quality endpoints,healthy_reference_choice
+94,88.0,Supp S04,Why the historical and strict protocols differ,guide:protocol
+95,89.0,Supp S04,Technical repeats and follow-up are different controls,healthy_immicro_replicate_controls
+96,90.0,Supp S04,Annotation agreement and difficult categories,healthy_annotation_agreement
+97,91.0,Supp S04,Depth matching changes precision and coverage,healthy_depth_matched_quality
+98,92.0,Supp S04,Detection and library size are separate diagnostics,healthy_detection_and_library
+99,93.0,Supp S05,Healthy cohort: LoRA cells cover the real ones,
+100,94.0,Supp S05,Healthy cohort: realism beyond precision/coverage,
+101,95.0,Supp S05,Healthy cohort: full vs LoRA vs condition encoder,
+102,96.0,Supp S05,Representative-dependent composition,
+103,97.0,Supp S05,Full vs LoRA vs cond.-encoder-only,
+104,98.0,Supp S06,"Pair-budget summary: composition, coverage and steps",
+105,99.0,Supp S06,Healthy cohort: LoRA rank and K,
+106,100.0,Supp S06,Healthy paired-data learning curves,
+107,101.0,Supp S06,"Time: days for scSemiProfiler, minutes for LoRA",
+108,102.0,Supp S06,5 donors: time to apply,
+109,103.0,Supp S06,"Pairs needed: composition, coverage, training steps",
+110,104.0,Supp S06,"LoRA rank: r = 4, 16, 64",
+111,105.0,Supp S06,K diversification: no gain,
+112,106.0,Supp S07,Five paired cohorts,
+113,107.0,Supp S07,Sepsis: exact-five reconstruction across five draws,disease_strict_sepsis_quality
+114,108.0,Supp S07,Sepsis: realism beyond precision and coverage,disease_strict_sepsis_realism
+115,109.0,Supp S07,Imperial: first completed exact-five comparison,disease_strict_imperial_quality
+116,110.0,Supp S07,5 donors: LoRA vs scSemiProfiler — cells,
+117,111.0,Supp S07,5 donors: realism beyond precision/coverage,
+118,112.0,Supp S07,"5 donors: precision/coverage at k = 3, 5, 10",
+119,113.0,Supp S07,5-fold: realism beyond precision/coverage,
+120,114.0,Supp S07,Same training pairs: LoRA vs scSemiProfiler,
+121,115.0,Supp S07,scSemiProfiler and the representative-copy baseline,
+122,,Supp S07,PBISC x1: completed evidence and remaining units,x1_priority:availability
+123,,Supp S07,Sepsis: full realism and depth controls,x1_priority:extras_ye2026_sepsis
+124,,Supp S07,Imperial ESKD: full realism and depth controls,x1_priority:extras_imperial_eskd
+125,,Supp S07,Giroux: full realism and depth controls,x1_priority:extras_giroux2022
+126,116.0,Supp S08,Sepsis: patient-aware disease-gene agreement,disease_strict_ye2026_sepsis_deg
+127,117.0,Supp S08,Imperial: patient-aware disease-gene agreement,disease_strict_imperial_eskd_deg
+128,118.0,Supp S08,Sepsis: fine-tuning variants,
+129,119.0,Supp S08,Pairs needed: donor and disease signal,
+130,120.0,Supp S08,"Cell-type deconvolution, same pairs: composition",
+131,121.0,Supp S08,Deconvolution: expression coverage limits the comparison,disease_deconv_expression_coverage
+132,122.0,Supp S08,"K=4,000 recipe: paired reconstruction quality",disease_k4000_quality
+133,123.0,Supp S08,"K=4,000 recipe: disease signal does not improve uniformly",disease_k4000_disease
+134,124.0,Supp S08,"K=4,000 recipe: calibration gains are selective",disease_k4000_calibration
+135,,Supp S08,Patient-aware DEG: same support and test samples,x1_priority:donor_deg
+136,,Supp S08,"Paired donor uncertainty: saved 2,000-bootstrap intervals",x1_priority:donor_intervals
+137,,Supp S08,Giroux: PBISC x1 donor uncertainty,x1_priority:individual_ci_giroux2022
+138,125.0,Supp S09,Influenza: response order versus response genes,disease_influenza_response
+139,126.0,Supp S09,COVID: calibration on a shared real-cell map,disease_covid_map
+140,127.0,Supp S09,COVID monocytes: direction after depth matching,disease_covid_direction
+141,128.0,Supp S09,"COVID monocytes: direction, magnitude and null",disease_covid_monocyte_direction_detail
+142,129.0,Supp S09,"COVID CD14 monocytes: direction, magnitude and null",disease_covid_cd14_direction_detail
+143,130.0,Supp S09,COVID: biological endpoints give a mixed verdict,disease_covid_endpoints
+144,131.0,Supp S10,Rare-state frequencies in observed cells,review:22
+145,132.0,Supp S10,Rare-state misses: rules 1–8,disease_rare_misses_1
+146,133.0,Supp S10,Rare-state misses: rules 9–15,disease_rare_misses_2
+147,134.0,Supp S10,Rare-state fidelity: zero-aware reanalysis,review:23
+148,135.0,Supp S10,Sepsis rare states across paired-data budgets,disease_sepsis_rare_support
+149,136.0,Supp S10,Sepsis rare-state direction across support budgets,disease_sepsis_rare_direction
+150,137.0,Supp S11,"Rare-state screen in DRESS, Kawasaki and SLE",disease_external_rare_screen
+151,138.0,Supp S11,Consequence: disease direction flips,
+152,139.0,Supp S11,C1Q: single-gene detection versus co-detection,disease_c1q_cross_disease
+153,140.0,Supp S11,"Behçet: K=4,000 changes direction more than identity",disease_behcet_k4000
+154,141.0,Supp S11,Behcet: one-gene and multi-gene C1Q rules,review:24
+155,142.0,Supp S12,Pairs from other cohorts do not transfer,
+156,143.0,Supp S12,Longer training on other cohorts: gap stays,
+157,144.0,Supp S12,"No pretraining: same cohort, from scratch",
+158,145.0,Supp S12,"No pretraining: other cohorts, from scratch",
+159,146.0,Supp S12,Rare states: from scratch vs pretrained,
+160,147.0,Supp S13,Appendix: scSemiProfiler without output threshold,
+161,148.0,Supp S14,Bulk-only exploration: calibration populations differ,disease_bulkonly_imperial_wave_shift
+162,149.0,Historical archive,Historical scoring and plans,
+163,150.0,Historical archive,Rare-state fractions: original analysis,
+164,151.0,Historical archive,Rare-state ranking: original analysis,
+165,152.0,Historical archive,Rare states: original five-fold analysis,
+166,153.0,Historical archive,Bulk-only cohort plan: original snapshot,
+167,154.0,Historical archive,Historical bulk-only search: no confirmed new finding,disease_bulkonly_verification_history
+168,155.0,Discussion,Discussion and outlook,
+169,156.0,Discussion,Next: other tissues from a few dozen pairs,