Multi-objective Optimization Algorithm For Molecular Docking

Molecular docking is a main method of structure-based drug design.In recent years, it has emerged as an important technology in the field. It is fundamentally an optimization problem of predicting the ligand-binding pose. As the objects during the optimization process, scoring functions estimate binding affinities between small ligands and proteins, and rank the compounds, playing an essential role in molecular docking. To date, most of available functions are designed based on single objective optimization method. They can generally be divided into the following three types, force-field-based, empirical, and the knowledge-based scoring functions. They focus on diverse aspects of ligand binding, and no one can satisfy the peculiarities of each target system. Despite the great efforts over the last decades, there is still an urgent need to improve the accuracy of scoring function. Therefore, an idea of consensus score strategy has emerged. Several methods have been developed based on it, scoring and ranking the compounds hierarchically with different scoring functions combined. The consensus score strategy arouses the inspiration of deriving our new scoring function with a multi-objective optimization method for molecular docking, and we hope this method may greatly improve the docking accuracy.Actually, there are more than twenty mathematical mutli-objective optimization techniques, and most of them compromise the objects to find pareto-optimal solutions from which the optimal design is chosen for a certain application. In this study, we introduce three commonly used scoring functions as the objectives, and integrate them into an equivalent function. While predicting the binding conformation, compounds are scored by all the scoring functions applied simultaneously. Considering more aspects of ligand binding with these different objectives, our method may result in more reasonable and robust binding poses.Tests of the derived scoring function against the GOLD test data set containing134protein-ligand complexes reveals a84.33%excellent result with a RMSD value below2.0A, which is much higher than the individual scoring methods used as objects in this study and another6well-known docking programs, indicating that our scoring function can be efficiently employed in molecular docking for ligand-binding pose prediction.