Interaction-Based Inductive Bias in Graph Neural Networks: Enhancing Protein-Ligand Binding Affinity Predictions from 3D Structures

Note

• Implementation of other baselines can be found on GIGN.
• This repository contains the source code for PLA prediction. For structure-based virtual screening (SBVS), please refer to our dedicated repository at EHIGN_SBVS on GitHub.

Dataset

All data used in this paper are publicly available at the following locations:

• PDBbind v2016 and v2019: pdbbind
• 2013 and 2016 core sets: casf

The preprocessed data can be downloaded from Graphs.

Requirements

dgl==0.9.0 networkx==2.5 numpy==1.19.2 pandas==1.1.5 pymol==0.1.0 rdkit==2022.3.5 scikit_learn==1.1.2 scipy==1.5.2 torch==1.10.2 tqdm==4.63.0 openbabel==3.3.1 (conda install -c conda-forge openbabel)

Alternatively, install the environment using the provided YAML file at ./environment.yaml.

Descriptions of Folders and Files

• ./data: Contains information about various datasets. Download and organize preprocessed datasets as described.
• ./config: Parameters used in EHIGN.
• ./log: Logger.
• ./model: Contains model checkpoints and training records.
• Scripts and Implementations: Various Python files implementing models, preprocessing, training, and testing.

Step-by-step Running

1. Model Training

• Download the preprocessed datasets and organize them in the ./data folder.
• Run python train.py.

2. Model Testing

• Run python test.py (modify file paths in the source code if necessary).

3. Process Raw Data

• Run a demo using provided examples:
  • python preprocess_complex.py
  • python graph_constructor.py
  • python train_example.py

4. Test the Trained Model in Other External Test Sets

• Organize the data like: -data   -external_test       -pdb_id         -pdb_id_ligand.mol2         -pdb_id_protein.pdb

• Execute the following commands:

  • python preprocess_complex.py
  • python graph_constructor.py
  • python test.py
  • (Modify file paths in the source code if necessary)

5. Cold Start Settings

• Use datasets found in the ./cold_start_data folder.
• Execute scripts train_random.py, train_scaffold.py, and train_sequence.py if the original training set has been processed.