zaixizhang / druggps_icml23

Implementation of ICML23 paper "Learning Subpocket Prototypes for Generalizable Structure-based Drug Design"

Learning Subpocket Prototypes for Generalizable Structure-based Drug Design

Official Pytorch implementation of ICML'23 paper "Learning Subpocket Prototypes for Generalizable Structure-based Drug Design" (https://arxiv.org/abs/2305.13997).

Previous work

• DrugGPS is based on our previous work on structure-based drug design: Molecule Generation For Target Protein Binding With Structural Motifs (ICLR 2023)
  • Code: https://github.com/zaixizhang/FLAG
  • Paper: https://openreview.net/forum?id=Rq13idF0F73

Install conda environment via conda yaml file

conda env create -f druggps_env.yaml
conda activate druggps_env

Datasets

Please refer to README.md in the data folder.

Preprocess dataset with BINANA

cd utils
python preprocess_data.py

Training

python train.py

Sampling

python sample.py

Generated Molecules for CrossDocked dataset

The generated molecules for 100 protein targets are stored in druggps.pt

The index file is test_index.pkl

Reference

@article{zhang2023learning,
  title={Learning Subpocket Prototypes for Generalizable Structure-based Drug Design},
  author={Zhang, Zaixi and Liu, Qi},
  journal={ICML},
  year={2023}
}

druggps_icml23's Issues

how to generate the molecules on my own protein pockets

hi, Zaixi,

Thanks for providing such interesting work!
Can you please give a instructions to generate compounds using the scripts in the repo to run inside the pockets that target my own specific proteins?
many thanks,

Best,
Sh-Y

FileNotFoundError:[Erron 2] No such file or directory:'. /data/crossdocked_pocket10_name2id.pt

Hi, Zaixi，
Thanks for providing so great work!
Maybe I'm missing some information, I don't have a file：'crossdocked_pocket10_name2id.pt’
Cause error "FileNotFoundError:[Erron 2] No such file or directory:'. /data/crossdocked_pocket10_name2id.pt'. I checked the code and found no problem, could you please provide some help?
Thank you!

how to convert .sdf or .pdb to .pdbqt?

Thanks for your wonderful work!
I'm more interested in how you converted .sdf and .pdb files to .pdbqt files? What tools are used? Can you provide some detailed explanation?
thanks again.

pocket representation issue

Hi, Zaixi,

Thank you very much for your work, but I have some questions about the representation of protein pockets in your article. As mentioned in Attachment A of your article, under Residue-level edge features, the residue features are calculated using adjacent residues i and j. However, the protein data in CrossDocked seems to have gaps,such as residue i and next is i+10. How did you handle the data with gaps?
many thanks

Best,
Zhongyu

Something about the generated molecules in the test dataset

Hi Zaixi,

Could you tell me druggps.pt is the file for the generated molecules for pockets in the test dataset?

Details About Pocket-based Clustered Split

Hi Zaixi, thanks for pointing out the problem with sequence-based clustered split!
I'm really interested in different ways of dataset split, but after going through PocketMatch document, I can't find the exact definition of similarity score mentioned in your paper (perhaps it is P-max_OP = C/B?) and how the clustering is done.
Could you please share more details about the splitting or hopefully the data processing script? Thanks in advance!