Multi-Objective Optimizations for Selections of Protein-Ligand Docking Poses Using Pareto Optima as a Consensus Score

In this study, new types of consensus scoring strategy using Pareto optimal concepts were proposed and tested. Our new strategy is using multiple scoring functions as objective functions, and the Pareto optima were selected. To evaluate the strategy, 458 protein-ligand complex structures whose three-dimensional structures were known were used. The reasonable docking poses having root mean square deviations against experimental structures less than 2.0 A were obtained from the Pareto optimal set for 75% of the test set. Although 10 types of scoring functions were used in this test, it was found that the most efficient number of scoring functions for a multi-objective strategy is four or five.