guaguabujianle / ehign Goto Github PK

Implementation of other baselines can be found in https://github.com/guaguabujianle/GIGN

All data used in this paper are publicly available and can be accessed here:

• PDBbind v2016 and v2019: http://www.pdbbind.org.cn/download.php
• 2013 and 2016 core sets: http://www.pdbbind.org.cn/casf.php
• DUD-E and DEKOIS2.0: https://zenodo.org/record/3975554#.Y1VoT44zaUk

You can download the preprocessed data from: https://drive.google.com/file/d/1oGUP4z7htNXyxTqx95HNSDLsaoxa3fX7/view?usp=share_link

dgl==0.9.0
networkx==2.5
numpy==1.19.2
pandas==1.1.5
pymol==0.1.0
rdkit==2022.3.5
scikit_learn==1.1.2
scipy==1.5.2
torch==1.10.2
tqdm==4.63.0
openbabel==3.3.1 (conda install -c conda-forge openbabel)

• ./data: This folder contains information about train, valid, test2013, test2016, and test2019 data sets. You should first download the preprocessed datasets from https://drive.google.com/file/d/1oGUP4z7htNXyxTqx95HNSDLsaoxa3fX7/view?usp=share_link, and put them into this folder and organize them as './data/train', './data/valid', './data/test2013/', './data/test2016/', and './data/test2019/'. We also provide a toy set with 50 examples to explain how to process raw data and train EHIGN model from scratch.
• ./config: Parameters used in EHIGN.
• ./log: Logger.
• ./model: A folder contains model checkpoints and records about the training processing.
• CIGConv.py: The implementation of CIGConv.
• NIGConv.py: The implementation of NIGConv.
• EHIGN.py: The implementation of EHIGN.
• HGC.py: The implementation of the heterogeneous graph neural network, where most of the contents are copied from the source code of dgl, but we have made some modifications so that it can process edge features.
• preprocess_complex.py: Prepare input complexes. The input ligand and protein should be .mol2 and .pdb formats, respectively. Proteins are first cropped around the ligand within 5 angstrom using pymol. Each protein and ligand is then processed into a protein-ligand complex using rdkit (i.e., a tuple that contains a ligand and a protein).
• graph_constructor.py: Convert protein-ligand complexes into heterogeneous graphs.
• train.py: Train EHIGN model.
• test.py, Test a trained model on 2013 core set, 2016 core set, and 2019 holdout sets and print the results.
• performance.py, Report the performance with three independent runs in mean (std) format.
• utils.py The File that includes useful tools for model training.

Firstly, download the preprocessed datasets from https://drive.google.com/file/d/1oGUP4z7htNXyxTqx95HNSDLsaoxa3fX7/view?usp=share_link, and put them into this folder and organize them as './data/train', './data/valid', './data/test2013/', './data/test2016/', and './data/test2019/'.
Secondly, run train.py using python train.py.

Run test.py using python test.py.
You may need to modify some file paths in the source code before running it.

We provide a demo to explain how to process the raw data. This demo use ./data/toy_examples.csv and ./data/toy_set/ as examples.
Firstly, run preprocess_complex.py using python preprocess_complex.py.
Secondly, run graph_constructor.py using python graph_constructor.py.
Thirdly, run train.py using python train_example.py.

Firstly, please organize the data as a structure similar to './data/toy_set' folder.
-data
  -external_test
      -pdb_id
        -pdb_id_ligand.mol2
        -pdb_id_protein.pdb
Secondly, run preprocess_complex.py using python preprocess_complex.py.
Thirdly, run graph_constructor.py using python graph_constructor.py.
Fourth, run test.py using python test.py.
You may need to modify some file paths in the source code before running it.