BiCoA-Net: Bidirectional Co-Attention Network

Model Description

BiCoA-Net predicts protein-ligand dissociation rate constants (k_off) using bidirectional co-attention mechanisms between protein and ligand representations.

Key Features:

• Predicts binding kinetics (k_off) for drug-target interactions
• Uses ESM-2 protein embeddings + MolFormer ligand embeddings
• Bidirectional co-attention fusion mechanism
• Trained on curated KineticX datasets

Training Details

• Optimizer: AdamW
• Learning Rate: 1e-4
• Batch Size: 32
• Epochs: 100
• Loss Function: MSE on log(k_off)

License

MIT License - Free for academic and commercial use.

Contact

For questions or issues, please open an issue on the GitHub repository or contact the authors.