EquiPocket: an E(3)-Equivariant Geometric Graph Neural Network for Ligand Binding Site Prediction

Welcome! These files are the source code for our work. (EquiPocket: an E(3)-Equivariant Geometric Graph Neural Network for Ligand Binding Site Prediction)

Files

• protein_feature.py: Extract the local geometric and global structure features in a protein, which needs MSMS to generate protein surface.

• models/

  • EquiPocket.py: Our proposed EquiPocket Framework.
  • surface_egnn.py: Our proposed Surface-Egnn model.
  • baseline_models.py: The implementation for some baseline models.
  • egnn_clean.py: The source code for "E(n) Equivariant Graph Neural Networks" from https://github.com/vgsatorras/egnn/tree/main/models.

• processed_data/

  • 5ei3_protein.pdb: A clean protein from the PDBbind dataset.
  • 5ei3_protein.pkl: The processed graph data for 5ei3_protein.

Dependency

• python: 3.7
• cuda: 11.6
• python packages: requirements.txt
• MSMS: https://ccsb.scripps.edu/msms/

Datasets

• scPDB: http://bioinfo-pharma.u-strasbg.fr/scPDB
• PDBbind: http://www.pdbbind.org.cn
• COACH420: https://github.com/rdk/p2rank-datasets/tree/master/coach420
• HOLO4K: https://github.com/rdk/p2rank-datasets/tree/master/holo4k