BA-Pred

Protein-Ligand Binding Affinity Prediction using Graph Neural Networks

High-performance protein-ligand binding affinity prediction model - 2nd place in CASP16 ligand affinity challenge

Features

• CASP16: 2nd place in the prestigious CASP16 ligand affinity prediction challenge
• High Accuracy: Graph Neural Network-based architecture for precise binding affinity prediction
• Research Ready: Pre-trained models ready for immediate use
• Easy Integration: Simple Python API and command-line interface
• Scalable: Batch processing for high-throughput screening

Quick Start

Installation

Choose your preferred installation method:

Option 1: Install from PyPI (Simplest)

pip install bapred

Option 2: Using Conda (Recommended for Development)

git clone https://github.com/eightmm/BA-Pred.git
cd BA-Pred
conda env create -f env.yaml
conda activate BAPred
pip install -e .

Option 3: From Source

git clone https://github.com/eightmm/BA-Pred.git
cd BA-Pred
pip install -r requirements.txt
pip install -e .

Run Your First Prediction

python run_inference.py -r example/1KLT.pdb -l example/ligands.sdf -o results.csv

That's it! Your binding affinity predictions will be saved in results.csv.

Usage Examples

Basic Usage

# Predict binding affinities
python run_inference.py -r example/1KLT.pdb -l example/ligands.sdf -o results.csv

Advanced Options

# Use CPU instead of GPU
python run_inference.py -r protein.pdb -l ligands.sdf -o results.csv --device cpu

# Custom batch size for memory optimization
python run_inference.py -r protein.pdb -l ligands.sdf -o results.csv --batch_size 64

# Limit CPU workers for data loading
python run_inference.py -r protein.pdb -l ligands.sdf -o results.csv --ncpu 8

# Specify custom model path
python run_inference.py -r protein.pdb -l ligands.sdf -o results.csv --model_path /path/to/model

Python API

from bapred.inference import inference

# Run prediction programmatically
inference(
    protein_pdb="example/1KLT.pdb",
    ligand_file="example/ligands.sdf",
    output="results.csv",
    batch_size=128,
    ncpu=4,
    model_path="bapred/weight",
    device="cuda"
)

Project Structure

BA-Pred/
├── bapred/                 # Main package
│   ├── data/               # Data processing modules
│   │   ├── atom_feature.py    # Atomic feature extraction
│   │   ├── data.py           # Dataset handling
│   │   └── utils.py          # Utility functions
│   ├── model/              # Neural network models
│   │   ├── GatedGCNLSPE.py   # Gated Graph Convolution
│   │   └── model.py          # Main model wrapper
│   ├── weight/             # Pre-trained weights
│   │   └── BAPred.pth        # Model checkpoint
│   └── inference.py       # Inference engine
├── example/               # Example files
│   ├── 1KLT.pdb             # Sample protein structure
│   └── ligands.sdf          # Sample ligand library
├── run_inference.py      # Easy-to-use script
├── requirements.txt      # Python dependencies
├── env.yaml             # Conda environment
└── README.md            # You are here!

Input/Output Formats

Input

• Protein: PDB format (.pdb)
• Ligands: SDF (.sdf), MOL2 (.mol2), or text file with paths (.txt)

Output

• CSV/TSV file with columns:
  • Name: Ligand identifier
  • pKd: Predicted binding affinity (pKd scale)
  • Kcal/mol: Binding energy in kcal/mol

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.

Citation

If you use BA-Pred in your research, please cite:

@software{bapred2024,
  title={BA-Pred: Protein-Ligand Binding Affinity Prediction using Graph Neural Networks},
  author={Jaemin Sim},
  year={2024},
  url={https://github.com/eightmm/BA-Pred}
}

Support

• Documentation: Check this README and code comments
• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

Made with care for the scientific community

Star us on GitHub if this project helped you!