SSGARL: Hybrid evolutionary computation and reinforcement learning for flexible ligand docking

Abstract

This paper presents and investigates the performance of a hybrid algorithm of steady-state genetic algorithm and reinforcement learning (SSGARL) in the problem of protein-ligand docking. The performance was measured in terms of the lowest found docking energy, the number of energy evaluation and the time taken to complete a docking task. Ten ligands of varying flexibility were chosen to bind with thermolysin to compare the performance of SSGARL and Iterated Local Search global optimizer's algorithm of AutoDock Vina. The results reveal that SSGARL finds the lowest docking energy, requires lesser number of energy evaluation and is faster in docking the highly flexible ligands.