期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2017
卷号：95
期号：21
页码：5685
出版社：Journal of Theoretical and Applied
摘要：Molecular docking is an important process in pharmaceutical research and drug design. It is used in screening libraries of small molecules or ligands to bind to a target protein changing its original biochemical properties forming new stable complex. In docking ligand to protein, the ligands pose, i.e. position, orientation and torsion angles, is translated, orientated, and the ligands torsion angles are rotated repeatedly to find an ideal site on the protein to bind. In this paper, Q-learning algorithm, a model-free reinforcement learning, with adaptive Kanerva Coding is used as the searching algorithm for protein-ligand docking problem. It evaluates the effectiveness of Q-learning algorithm and the different settings for the parameters of reinforcement learning. A popular docking tool called AutoDock Vina was used to find the ligands goal pose. The effectiveness of the agent is measured by the success of finding the goals. The proposed agent managed to match and finds better pose than AutoDock Vina in medium to large size ligands.
