📢 ResGen: A Pocket-aware 3D Molecular Generation Model Based on Parallel Multi-scale Modeling

ResGen: A Pocket-aware 3D Molecular Generation Model Based on Parallel Multi-scale Modeling

ResGen is the newly developed method for 3D pocket-aware molecular generation.

Environment

Install via conda yaml file (cuda 11.3)

mamba env create -f resgen.yml
mamba activate resgen 

Install manually

(we recommend using mamba instead of conda, if you're old school, just change the mamba with conda)

mamba create -n resgen python=3.8
mamba install pytorch==1.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
mamba install pyg -c pyg
mamba install -c conda-forge rdkit
mamba install biopython -c conda-forge
mamba install pyyaml easydict python-lmdb -c conda-forge

Data

The main data for training is CrossDock2020, which is utilized by most of the methods.

Note: data is only necessary for training. For use-only mode, please directly check the generation part.

(Optional) Download the data from the original source

wget https://bits.csb.pitt.edu/files/crossdock2020/CrossDocked2020_v1.1.tgz -P data/crossdock2020/
tar -C data/crossdock2020/ -xzf data/crossdock2020/CrossDocked2020_v1.1.tgz
wget https://bits.csb.pitt.edu/files/it2_tt_0_lowrmsd_mols_train0_fixed.types -P data/crossdock2020/
wget https://bits.csb.pitt.edu/files/it2_tt_0_lowrmsd_mols_test0_fixed.types -P data/crossdock2020/

The storage size of original crossdock2020 is 50 GB, hard to download and unzip. You can skip to the Approach 1 or Approach 2 for training preparation.

Approach 1: Download the Pocket Data for Processing

You can download the processed data from this link. This is the processed version of original files, which is processed by Luoshi Tong.

Note: index.pkl, split_by_name.pt. are automatically downloaded with the SurfGen code. index.pkl saves the information of each protein-ligand pair, while split_by_name.pt save the train-test split of the dataset.

tar -xzvf crossdocked_pocket10.tar.gz
python process_data.py --raw_data ./data/crossdocked_pocket10 

Approach 2: Download the Processed Data

or you can download the processed data lmdb, key, and name2id.

Generation

The trained model's parameters could be downloaded here.

python gen.py --pdb_file ./examples/4iiy.pdb --sdf_file ./examples/4iiy_ligand.sdf --outdir ./examples

You can also follow the guide at generation/generation.ipynb

We provide the pdbid-14gs as the example

Training

The training process is released as train.py, the following command is an example of how to train a model.

python train.py --config ./configs/train_res.yml --logdir logs

Acknowledge

This project draws in part from GraphBP and Pocket2Mol, supported by GPL-v3 License and MIT License. Thanks for their great work and code, hope readers of interest could check their work, too.