EISG

TU benchmark：D&D, MUTAG, PROTEINS.

DrugOOD：IC50/EC50-size/scaffold/assay.

Folder Specification

• data/: containing the data for training, including the preprocess result and substructure information merging scripts.
• config/: configuration for model and model training.
• teain_eval.py: the script to train our algorithm.
• figure/: containing the visualization of QED dataset.
• models/: containing backbone definition for our method.
• losses.py: containing the loss definition.
• main.py: the script to run.
• vis.py: the script to visualization.

Package Dependency

python: 3.9.19
pytorch: 1.12.1          # With CUDA 11.3 support
torch-geometric: 2.6.0    # PyTorch Geometric (PyG)
torchvision: 0.13.1       # Computer vision extension
torchaudio: 0.12.1        # Audio processing extension
drugood: 0.0.1            # Drug discovery toolkit
rdkit: 2023.9.5           # Cheminformatics toolkit (latest stable)
numpy: 1.26.4             # Numerical computing foundation
cudatoolkit: 11.3.1       # CUDA toolkit
dgl-cu110: 0.6.1          # Deep Graph Library (CUDA 11 compatible)
datasets: 2.20.0          # HuggingFace dataset handling
transformers: 4.43.4      # Pretrained models library
scikit-learn: 1.2.2       # Machine learning toolkit
pandas: 2.2.2             # Data analysis framework

Data Generation

• DruOOD ：The first step is to generate the original dataset from CHEMBL database. As for the detailed process or operation, please refer to the DrugOOD repository. The generated files should be put into folder or respectively.
• json``DrugOOD/data/ic50``DrugOOD/data/ec50
• TU ：auto download.

Run the Code

run main.py