The official implementation of paper: FROM SINGLETON IN TO PAIRWISE: GENERATING LIG- AND CONFORMATION WITH LIGAND-TARGET INTERACTIONS

Our model runs on Tesla A100 40G GPUs, you can create the environment by conda:

# Clone the environment
conda env create -f environment.yml

The original PDBBind-2020 dataset can be derived at http://www.pdbbind.org.cn/index.php?action=showall We provide our processed version of training, validation, and testing dataset at: GoogleDrive Folders. After downloding the dataset, it should be put into the folder path as specified in the datasets/

The trained checkpoints for both score function and energy function are saved in: GoogleDrive Folders s_theta.pt for score function, gap.pt, energy.pt and charge.pt After downloding the checkpoints, it should be put into the folder path as specified in the logs/

All hyper-parameters and training details are provided in config files (./configs/pdbbind_default.yml), you can tune the parameters. You can train the score function model with the following commands:

bash train_ddp.sh

You can train the energy function model with the following commands:

bash train_g_phi.sh

The model checkpoints will be saved in /logs folder

You can use the following commands to generate samples:

bash sample.sh

The generated mols will be saved in /samples folder

The evaluation contains RMSD and ligand RMSD. After evaluation, the ligand files will be saved in .mol format which can be opened by pymol. To run the evaluation, run

python eval.py