Reinforcement Learning Based Approach for Ligand Pose Prediction

Abstract

Identification of the potential binding site and the correct ligand pose are two crucial steps among the various steps in protein ligand interaction for a novel or known target. Currently most of the deep learning methods work on protein ligand pocket datasets for various predictions. In this study, we propose a reinforcement learning (RL) based method for predicting the optimized ligand pose where the RL agent also identifies the binding site based on its training. In order to apply various reinforcement learning techniques, we suggest a novel approach to represent the protein ligand complex using graph CNN which would help utilize both atomic and spatial features. To the best of our knowledge, this is the first time an RL based approach has been put forward for predicting optimized ligand pose.