InteractionGraphNet: a Novel and Efficient Deep Graph Representation Learning Framework for Accurate Protein-Ligand Interaction Prediction and Large-scale Structure-based Virtual Screening.
# software
chimera v1.10.1 (https://www.cgl.ucsf.edu/chimera/cgi-bin/secure/chimera-get.py?file=linux_x86_64/chimera-1.10.1-linux_x86_64.bin)
openbabel (https://open-babel.readthedocs.io/en/latest/Installation/install.html#compiling-open-babel)
# python
python 3.6.5
# environment to reproduce
conda create --prefix xxx --file ./env/ign_needs_conda.txt
pip install -r ./env/ign_needs_pip.txt
nohup python -u ./codes/ign_train.py --gpuid 0 --epochs 5 --repetitions 3 --lr 0.0001 --l2 0.000001 --dropout 0.1 > ./log/toy_example.log 2>&1 &
We added only about data of 200 toy samples or so in the data folder due to the large data size to explain how to train IGN model. Each sample is saved in a pickle file and it consists of two rdkit objects of a ligand and protein pocket prepared by chimera software.
We use the well-trained IGN model to predict the binding affinity of complexes generated from docking program
# step 1
# mol file format conversion
# the mol2 files in ./examples/mol2_files folder are the conformers generated from docking program
python3 ./codes/mol2tosdf.py --mol2path ./examples/mol2_files --sdfpath ./examples/sdf_files --num_process 12
# step 2
# select residues using chimera for each ligand/protein pair
python3 ./codes/select_residues.py --proteinfile ./examples/protein_6exw.pdb --sdfpath ./examples/sdf_files --finalpath ./examples/ign_input --num_process 12
# step 3
# prediction
python3 ./codes/prediction.py --cpu True --num_process 12 --input_path ./examples/ign_input