README.md

---
tags:
- dnabert
- bacteria
- kmer
- translation-initiation-site
- sequence-modeling
- DNA
library_name: transformers
---

# BacteriaTIS-DNABERT-K6-89M

This model, `BacteriaTIS-DNABERT-K6-89M`, is a **DNA sequence classifier** based on **DNABERT** trained for **Translation Initiation Site (TIS) classification** in bacterial genomes. It operates on **6-mer tokenized sequences** derived from a **60 bp window (30 bp upstream + 30 bp downstream)** around the TIS. The model was fine-tuned using **89M trainable parameters**.

## Model Details
- **Base Model:** DNABERT
- **Task:** Translation Initiation Site (TIS) Classification
- **K-mer Size:** 6
- **Input Sequence Window:** 60 bp (30 upstream + 30 downstream) of TIS site in ORF sequence
- **Number of Trainable Parameters:** 89M
- **Max Sequence Length:** 512
- **Precision Used:** AMP (Automatic Mixed Precision)

---


### **Install Dependencies**
Ensure you have `transformers` and `torch` installed:
```bash
pip install torch transformers
```

### **Load Model & Tokenizer**
```python
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

# Load Model
model_checkpoint = "Genereux-akotenou/BacteriaTIS-DNABERT-K6-89M"
model = AutoModelForSequenceClassification.from_pretrained(model_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
```

### **Inference Example**
To classify a TIS, extract a 60 bp sequence window (30 bp upstream + 30 bp downstream) of the TIS codon site and convert it to 6-mers:
```python
def generate_kmer(sequence: str, k: int, overlap: int = 1):
    """Generate k-mer encoding from DNA sequence."""
    return " ".join([sequence[j:j+k] for j in range(0, len(sequence) - k + 1, overlap)])

# Example TIS-centered sequence (60 bp window)
sequence = "AGAACCAGCCGGAGACCTCCTGCTCGTACATGAAAGGCTCGAGCAGCCGGGCGAGGGCGG"
seq_kmer = generate_kmer(sequence, k=6)
```

### **Run Model**
```python
# Tokenize input
inputs = tokenizer(
  seq_kmer,
  return_tensors="pt",
  max_length=tokenizer.model_max_length,
  padding="max_length",
  truncation=True
)

# Run inference
with torch.no_grad():
  outputs = model(**inputs)
  logits = outputs.logits
  predicted_class = torch.argmax(logits, dim=-1).item()
```

<!-- ### **Citation**
If you use this model in your research, please cite:
```tex
@article{paper title,
  title={DNABERT for Bacterial Translation Initiation Site Classification},
  author={Genereux Akotenou, et al.},
  journal={Hugging Face Model Hub},
  year={2024}
}
``` -->