Prediction of protein-ligand binding affinities using a graph convolutional neural network model and improving the performance of the model by data augmentation techniques

Abstract

Prediction of protein-ligand interactions is a critical step during the initial phase of drug discovery. I have developed a novel deep-learning-based prediction model based on a graph convolutional neural network estimating the protein-ligand binding affinity. Graph convolutional neural networks extract features more efficiently with reduced the computational time and resources that are normally required by the traditional convolutional neural network models. The protein-ligand complex is described as a graph that can be constructed with nodes and edges. The model utilizes graph convolution using multiple adjacency matrices whose entries are affected by distances, and a feature matrix that describes the molecular properties of the atoms. The model for protein-ligand binding affinities was tested on the PDBbind datasets and proved the accuracy and the efficiency of the graph convolution. The computational efficiency of graph convolutional neural networks enables data augmentation with docking simulation. I found that data augmentation with docking simulation data could improve the prediction accuracy when the generated structures are accurate and the number of docking structure is sufficient. The high prediction performance and speed of the graph convolutional neural network model suggest that such networks can serve as valuable tools in drug discovery.