Energy-based Generative Models for Target-specific Drug Discovery

A Pytorch implementation of Target-specific Generation of Molecules (TagMol). This library refers to the following source code.

• yongqyu/MolGAN-pytorch
• jundeli/quantum-gan

For details see Energy-based Generative Models for Target-specific Drug Discovery by Junde Li, Collin Beaudoin, and Swaroop Ghosh.

Dependencies

• python>=3.5
• pytorch>=0.4.1: https://pytorch.org
• frechetdist

Structure

• data: should contain your protein-ligand complex datasets. Download PDBbind dataset from: https://pennstateoffice365-my.sharepoint.com/:f:/g/personal/jul1512_psu_edu/ErM3Iuz_OjNMnHTsZyWVhGQBtLCakhUin4bqMShQWXEpKA?e=nRxKQy
• models: Class for Models.

Training

python main.py

This main file was used for running TagMol experiments with GCN and GAT backends. The file data_dump.py could be executed first in order to run experiments faster with a preprocessed dataset.

python cgan_tagmol.py

This is a simplified cgan version of tagmol exluding the energy network. It supports generating target-specific molecules but not evaluating the relative binding affinity between the protein target and ligand.

Below are some generated molecules:

Citation

@article{li2022energy,
  title={Energy-based Generative Models for Target-specific Drug Discovery},
  author={Li, Junde and Beaudoin, Collin and Ghosh, Swaroop},
  journal={arXiv preprint arXiv:2212.02404},
  year={2022}
}