peptcrnet 1.0.3

A Deep Learning Framework for TCR-Peptide Recognition Prediction

PepTCR-Net: Deep Learning for TCR-Peptide Recognition Prediction

PepTCR-Net is a state-of-the-art deep learning framework for predicting T-cell receptor (TCR) recognition of peptide antigens. It combines advanced neural network architectures with comprehensive feature engineering to achieve high-accuracy predictions with uncertainty quantification.

🌟 Key Features

• Multi-modal Integration: Seamlessly combines sequence, categorical, and network-based features
• Advanced Embeddings: Utilizes autoencoders, position encoding, and Atchley factors for sequence representation
• Bayesian Neural Networks: Provides uncertainty quantification for predictions
• Comprehensive Pipeline: End-to-end solution from data preprocessing to model deployment
• Flexible Architecture: Modular design allows easy customization and extension
• Class Imbalance Handling: Built-in support for imbalanced datasets
• Rich Visualizations: Extensive plotting utilities for model interpretation

🚀 Quick Start

Install from PyPI:

pip install peptcrnet

For notebook demos, also install the optional notebook dependencies: pip install peptcrnet[notebooks].

Run the Complete Demo (Easiest!)

# One-click demo launcher
./run_demo.sh

This launches the complete Scenario 17 demo using all features!

Installation

From PyPI (recommended)

pip install peptcrnet

For notebooks and demos: pip install peptcrnet[notebooks]

Requirements: Python 3.8–3.12 (Python 3.13 is not supported due to dependency constraints).

From source (development)

git clone https://github.com/mlizhangx/Pep-TCRNet.git
cd Pep-TCR-Net
pip install -e ".[notebooks]"

# Run the demo
jupyter notebook DEMO_Complete_Pipeline.ipynb

Basic Usage

import peptcrnet
from peptcrnet import PepTCRNetPipeline

# Initialize pipeline
pipeline = PepTCRNetPipeline(data_path='your_data.csv')

# Load and prepare data
pipeline.load_data()
pipeline.split_data(test_size=0.2, val_size=0.1)

# Prepare features
pipeline.prepare_features(feature_types=['sequences', 'categorical'])

# Train model
history = pipeline.train(epochs=100, batch_size=128)

# Evaluate with uncertainty
results = pipeline.evaluate_with_uncertainty(n_samples=200)

# Make predictions
predictions = pipeline.predict(new_data)

📊 Data Format

PepTCR-Net expects input data in CSV format with the following columns:

Column	Description	Example
CDR3	TCR CDR3β sequence	CASSRGQGNEQFF
Peptide	Peptide sequence or class label	GILGFVFTL
V	V gene segment	TRBV7-2
J	J gene segment	TRBJ2-1
HLA-A	HLA-A allele	A*02:01
HLA-B	HLA-B allele	B*07:02
HLA-C	HLA-C allele	C*07:01

🧪 Demo Notebook

Try our interactive demo notebook to see PepTCR-Net in action:

jupyter notebook demo_pipeline.ipynb

The demo includes: - Sample data generation - Step-by-step pipeline walkthrough - Model training and evaluation - Uncertainty quantification - Visualization examples

📚 Documentation

Pipeline Components

1. Data Loading and Preprocessing

from peptcrnet.data import DataLoader

loader = DataLoader('data.csv', atchley_path='atchley_factors.txt')
stats = loader.get_summary_stats()
splits = loader.split_data()

2. Feature Engineering

from peptcrnet.embeddings import SequenceEmbedder, CategoricalEmbedder

# Sequence embeddings
seq_embedder = SequenceEmbedder(atchley_factors, max_length=30)
tcr_embeddings = seq_embedder.encode_sequences(tcr_sequences)

# Categorical embeddings
cat_embedder = CategoricalEmbedder()
cat_embeddings = cat_embedder.encode_features(categorical_data)

3. Model Training

from peptcrnet.models import BayesianClassifier

model = BayesianClassifier(
    input_shapes={'sequences': (100,), 'categorical': (50,)},
    num_classes=5,
    hidden_dims=[512, 256, 64]
)

history = model.train(X_train, y_train, X_val, y_val)

4. Evaluation and Visualization

from peptcrnet.evaluation import ModelEvaluator
from peptcrnet.visualization import plot_confusion_matrix, plot_roc_curves

evaluator = ModelEvaluator()
metrics = evaluator.compute_metrics(y_true, y_pred, y_proba)

plot_confusion_matrix(y_true, y_pred)
plot_roc_curves(y_true, y_proba)

⚙️ Configuration

PepTCR-Net uses a centralized configuration system:

from peptcrnet import config

# Access configuration
print(config.ModelParams.MAX_TCR_LENGTH)
print(config.TrainingParams.BATCH_SIZE)

# Save configuration
config.save_config('my_config.json')

# Load configuration
config.load_config('my_config.json')

🔬 Advanced Features

Uncertainty Quantification

PepTCR-Net provides Bayesian uncertainty estimation:

# Multiple forward passes for uncertainty
predictions, uncertainty = pipeline.predict_with_uncertainty(
    test_data,
    n_samples=200
)

# Identify high-confidence predictions
high_confidence_mask = uncertainty < threshold

Custom Feature Combinations

Experiment with different feature combinations:

# Define feature cases
feature_cases = {
    1: ['TCR'],
    2: ['TCR', 'Peptide'],
    3: ['TCR', 'Peptide', 'HLA'],
    4: ['TCR', 'Peptide', 'HLA', 'VJ', 'Network']
}

# Train with specific features
pipeline.prepare_features(feature_types=feature_cases[3])

Model Persistence

Save and load trained models:

# Save complete pipeline
pipeline.save_pipeline('output_dir/')

# Load saved pipeline
new_pipeline = PepTCRNetPipeline()
new_pipeline.load_pipeline('output_dir/')

📈 Performance

PepTCR-Net achieves state-of-the-art performance on TCR-peptide binding prediction:

• Accuracy: Up to 95% on benchmark datasets
• AUC-ROC: 0.90 for multi-class classification
• Uncertainty Calibration: Well-calibrated confidence scores

🤝 Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

# Fork the repository
# Create your feature branch
git checkout -b feature/amazing-feature

# Commit your changes
git commit -m 'Add amazing feature'

# Push to the branch
git push origin feature/amazing-feature

# Open a Pull Request

📝 Citation

If you use PepTCR-Net in your research, please cite:

@article{le2025peptcrnet,
  title={PepTCR-Net: prediction of multi-class antigen peptides by T-cell receptor sequences with deep learning},
  author={Le, Phi and Ung, Leah and Yang, Hai and Huang, Anwen and He, Tao and Bruno, Peter and Oh, David Y and Keenan, Bridget P and Zhang, Li},
  journal={Briefings in Bioinformatics},
  volume={26},
  number={4},
  pages={bbaf351},
  year={2025},
  doi={10.1093/bib/bbaf351},
  url={https://doi.org/10.1093/bib/bbaf351}
}

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

📮 Contact

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: [e-mail]