EpitopeGen

EpitopeGen is a deep learning model that predicts cognate epitope sequences from T-cell receptor (TCR) sequences. It helps functionally annotate TCRs in single-cell TCR sequencing data by generating potential binding epitopes and identifying phenotype associations.

Installation

pip install epitopegen

Quick Start

from epitopegen import EpitopeGenPredictor

# Initialize predictor
predictor = EpitopeGenPredictor()

# Predict epitopes for TCR sequences
tcrs = ["CASIPEGGRETQYF", "CAVRATGTASKLTF"]
results = predictor.predict(tcrs)

Basic Usage

1. Prepare Input Data

Create a CSV file with TCR sequences:

text,label
CASIPEGGRETQYF,ZZZZZ
CAVRATGTASKLTF,ZZZZZ

2. Run Predictions

import pandas as pd
from epitopegen import EpitopeGenPredictor

# Initialize predictor
predictor = EpitopeGenPredictor()

# Read TCR sequences
tcrs = pd.read_csv("input.csv")["text"].tolist()

# Generate predictions
results = predictor.predict(
    tcr_sequences=tcrs,
    output_path="predictions.csv",
    top_k=50  # num of epitopes to generate
)

3. Annotate Phenotypes

from epitopegen import EpitopeAnnotator

# Initialize annotator with reference database
annotator = EpitopeAnnotator("epitopes_db.csv")

# Annotate predictions
results = annotator.annotate(
    predictions_df=results,
    method='substring',
    output_path="annotations.csv"
)

Key Features

• Generate potential epitope sequences for TCRs
• Support for multiple model checkpoints
• Phenotype annotation using reference databases
• Ensemble predictions for robust results
• Built-in analysis tools

Resources

• Detailed Tutorial
• API Reference
• Model Checkpoints
• Reference Databases
• Development Resources

Requirements

• Python ≥ 3.8
• PyTorch ≥ 1.12
• transformers ≥ 4.39.0
• pandas ≥ 1.5.0

Citation

If you use EpitopeGen in your research, please cite:

@article{epitopegen2024,
  title={Generating cognate epitope sequences of T-cell receptors with a generative transformer},
  year={2024}
}

License

MIT License

Support

For questions and issues:

• Open an issue on GitHub
• Contact: minukma@cs.ubc.ca