Awesome

MolRL-MGPT

This is the code repository for our paper published in NeurIPS 2023: De novo Drug Design using Reinforcement Learning with Multiple GPT Agents.

Dependencies

pytorch==1.12.1
rdkit==2020.03
tqdm
tensorboard
multiprocessing
PyTDC
openbabel

Dataset & Docking

Following Chemformer, we use a filtered ZINC dataset containing 100M SMILES. The files are available at MolecularAI/Chemformer.

The ChEMBL dataset is available at ChEMBL.

The SMILES vocabulary and protein structures can be found in data/.

Quick Vina 2 is available at QuickVina.

Pre-training

python codes/pretrain.py 

Multi-agent Reinforcement Learning

GuacaMol benchmark

python codes/MARL.py --task_id 0

SARS-COV-2 protein targets

python codes/MARL.py --oracle docking_PLPro_7JIR_mpo
python codes/MARL.py --oracle docking_RdRp_mpo

Citation

@article{hu2024novo,
  title={De novo Drug Design using Reinforcement Learning with Multiple GPT Agents},
  author={Hu, Xiuyuan and Liu, Guoqing and Zhao, Yang and Zhang, Hao},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2024}
}