BBERT-models

BBERT Pre-trained Models

Pre-trained models for BBERT - BERT for Bacterial DNA Classification.

Models Included

1. BBERT Transformer (bbert_checkpoint-32500/)

• Main BERT-based model trained on bacterial DNA sequences
• Hidden size: 768
• Trained on diverse bacterial genomes

2. Bacterial Classifier (bacterial_classifier/epoch_80.pt)

• Binary classifier for bacterial vs. non-bacterial sequences
• Input: BBERT embeddings (768-dim)
• Trained for 80 epochs on 3.9M sequences

3. Reading Frame Classifier (frame_classifier/classifier_model_2000K_37e.pth)

• 6-way classifier for reading frame prediction
• Frames: +1, +2, +3, -1, -2, -3
• Trained for 37 epochs on 2M sequences

4. Coding Sequence Classifier (coding_classifier/epoch_46.pt)

• Binary classifier for coding vs. non-coding sequences
• Trained for 46 epochs on 3.9M sequences

Usage

These models are automatically downloaded when using BBERT:

```bash

First time setup

pip install bbert # or clone from GitHub python source/download_models.py

Then use normally

python bbert.py your_sequences.fasta --output_dir results ```

Citation

If you use BBERT, please cite: [Add your citation here]

License

MIT License - see LICENSE file for details