An Automated Ligand Evolution System using Bayesian Optimization Algorithm∗

Abstract: Ligand docking checks whether a drug chemical called ligand matches the target receptor protein of human organ or not. Docking by computer simulation is becoming popular in drug design process to reduce cost and time of the chemical experiments. This paper presents a novel approach generating optimal ligand structures from scratch based on de novo ligand design approach employing Bayesian optimization algorithm to realize an automated design of drug and other chemical structures. The proposed approach searches an optimal structure of ligand that minimizes bond energy to the receptor protein, and the structure of ligand is generated by adding small fragments of molecules to the base structure. The decision of adding fragments are controlled by Bayesian optimization algorithm which is considered as a promising approach in probabilistic model-building genetic algo-rithms. We have built a system that automatically generates an optimal structure of ligand, and through numerical experiments performed on a PC cluster, we show the effectiveness of our approach compared to the conventional approach using classical genetic algorithms. Key–Words: automated drug design, ligand docking, screening, de novo ligand design approach, probabilistic model-building genetic algorithms, estimation of distribution of algorithms, Bayesian optimization algorithms