OnionNet-2

OnionNet-2 is constructed based on convolutional neural network (CNN) to predict the protein-ligand binding affinity. One of the greatest advantages of OnionNet-2 is that it can achieve higher accuracy at a lower computational cost. When taking CASF-2016 and CASF-2013 as benchmark, OnionNet-2 shows strong scoring power.

Contact

Prof. Weifeng Li, Shandong University, lwf@sdu.edu.cn

Zechen Wang, Shandong University, zechenwang@mail.sdu.edu.cn

Citation

Wang Z, Zheng L, Liu Y, et al. OnionNet-2: A Convolutional Neural Network Model for Predicting Protein-Ligand Binding Affinity based on Residue-Atom Contacting Shells. Front Chem 2021, 9, 753002.

Installation

First, create a conda environment and install some necessary packages for running OnionNet-2.

conda create -n OnionNet-2 python=3.8
conda activate OnionNet-2

conda install numpy
conda install pandas
conda install scikit-learn=0.22.1
conda install -c openbabel openbabel
conda install -c conda-forge mdtraj

Or, you can also use pip to install above packages. For example,

pip install tensorflow-gpu==2.3

Usage

1. Prepare the PDB file containing the 3D structure of protein-ligand complexes.

In the samples/prepare_complexes directory, we provide two scripts to conveniently prepare the PDB file containing 3D structure of the protein-ligand complex.

# Specify the relative path or absolute path of the working directory correctly. For example, 
bash prepare_PDB_file.sh ..

2. Generate the residue-atom contact features.

python generate_features.py -h
python generate_features.py -shells N -inp inputs_complexes.dat -out output_featurs.csv

The input file (inputs_complexes.dat) contains the path of the protein-ligand complexes pdb files, for example

1a30/1a30_complex.pdb
1bcu/1bcu_complex.pdb

3. Train the convolutional neural network.

# The features and the real pKa are concated in a common file as the input of the training process. We can execute this process with a script.   
python concat_features_pKa.py -inp_features output_features.csv -inp_true all_complexes_pKa.csv -out output_features_pKa.csv

# The training process will output 3 files, the default are "logfile", "bestmodel.h5" and "train_scaler.scaler"， of which the latter two will be the input of the prediction process. 
python train.py -h
python train.py -train_file train_features_pKa.csv -valid_file valid_features_pKa.csv -shape 84 124 1

4. Predict the protein-ligand binding affinity.

python predict_pKa.py -h
python predict_pKa.py -scaler train_scaler.scaler -model bestmodel.h5 -inp input_features.csv -out predicted_pKa.csv

aolgac/OnionNet-2