README.md

metadata

license: mit
language:
  - en
tags:
  - biology
  - genomics
  - viral-classification
  - transformer
  - distilbert

Viro-BERT

Whole-genome Transformer framework for viral pathogen classification across 14 viral classes using Chunk-and-Vote Late Fusion Consensus.

---

Requirements

pip install torch transformers biopython

Tested on Python 3.8+ and PyTorch 1.13+ (supports CPU and CUDA).

---

Files in Repository

• sota_chunk_model.pth — Trained PyTorch model weights (DistilBERT backbone, 6 layers, 12 heads, hidden dim 768).
• config.json — 130-token 3-mer vocabulary, hyperparameters, and 14 target class mappings.
• predict.py — Standalone inference script for FASTA sequence classification.

---

Command-Line Usage (Python)

Classify viral sequences from a FASTA file:

python predict.py --fasta sample.fasta

Optional arguments:

python predict.py --fasta sample.fasta --weights sota_chunk_model.pth --config config.json --top_k 3 --cpu

---

Execution via R

Run directly from an R environment via system command:

system("python predict.py --fasta sample.fasta")

Or programmatically via the reticulate interface:

library(reticulate)
py_run_file("predict.py")

---

Target Viral Classes (14)

• Dengue Virus (Dengue)
• Enterovirus A (EnteroA)
• Enterovirus B (EnteroB)
• Hepatitis B Virus (HBV)
• Hepatitis C Virus (HCV)
• Human Papillomavirus (HPV)
• Human Alphaherpesvirus 1 (HSV-1)
• Influenza A Virus (IAV)
• Influenza B Virus (IBV)
• Influenza C Virus (ICV)
• Monkeypox Virus (MPV)
• Severe Fever with Thrombocytopenia Syndrome Virus (SFTS)
• West Nile Virus (WNV)
• Zika Virus (Zika)

---

License

Open access under the MIT License.