DiffInt: A Diffusion Model for Structure-Based Drug Design with Explicit Hydrogen Bond Interaction Guidance

This repository provides the source code and pretrained models for the paper.

DiffInt is a diffusion-based generative model that explicitly incorporates hydrogen bond interactions for structure-based drug design. It introduces interaction particles to guide ligand generation within protein binding pockets.

Dependencies

Cuda environment

Software	Version
CUDA	11.8
cudnn	8.9.7

conda environment

conda env create -n Int-env -f environment.yml

Software	Version
Python	3.10.4
numpy	1.22.3
PyTorch	2.0.1
PyTorch cuda	11.8
Torchvision	0.15.2
Torchaudio	2.0.2
PyTorch Scatter	2.1.1
PyTorch Lightning	1.7.4
RDKit	2022.03.2
WandB	0.13.1
BioPython	1.79
imageio	2.21.2
SciPy	1.7.3
OpenBabel	3.1.1
ODDT	0.7

Data download

Download the training, validation and test datasets: Data

tar xvzf DiffInt_crossdock_data.tar.gz

Data construction by yourself

(You don't need to construct data by yourself.) Download and extract the dataset as described by the authors of Pocket2Mol.
Download the dataset archive crossdocked_pocket10.tar.gz and the split file split_by_name.pt to data directory.

.
├── data
│   ├── DiffInt_crossdock_data.tar.gz
│   └── split_by_name.pt

Extract the TAR archive

tar -xzvf crossdocked_pocket10.tar.gz

data preparation step 1

python process_crossdock.py /data/directory/path/ --outdir /output/directory/path/

For example

python process_crossdock.py data/ --outdir data/crossdocked_ca/

data preparation step 2: add hydrogen bonds information

python hbond_double.py --data_dir /step_1/directory/path/ --out_dir /step_2/directory/path/ --pdb_dir /pdb_data/directory/path/

For example

python hbond_double.py --data_dir data/crossdock_ca/ --out_dir data/crossdocked_ca_Int/ --pdb_dir data/crossdocked_pocket10/

Training

python -u train.py --config config/DiffInt_ca_double.yml

Molecule generation

Generation of 100 ligand molecules for 100 protein pockets.

python test_npz.py --checkpoint checkpoint_file --test_dir /data/directory/path/ --outdir /out/directory/path/

For example

python test_npz.py --checkpoint checkpoints/best_model.ckpt --test_dir DiffInt_crossdock_data/ --outdir sample

Generated molecules used in the paper are example/DiffInt_generated_molecules.tar.gz

Generate 100 ligand molecules for one pocket

python test_single.py --checkpoint checkpoint_file --outdir /out/directory/path/ --pdb /pdb/file/path/ --sdf /sdf/file/path/

Or you can use Google Colabratory (This notebook has been confirmed to work on May 28, 2025.)

.
├── colab
│   └── DiffInt_generate.ipynb