Optimizing the Edge Weights in Optimal Assignment Methods for Virtual Screening with Particle Swarm Optimization

Abstract

Ligand-based virtual screening experiments are an important task in the early drug discovery stage. In such an experiment, a chemical database is searched for molecules with similar properties to a given query molecule. The optimal assignment approach for chemical graphs has proven to be a successful method for various cheminformatic tasks, such as virtual screening. The optimal assignment approach assumes all atoms of a query molecule to have the same importance. This assumption is not realistic in a virtual screening for ligands against a specific protein target. In this study, we propose an extension of the optimal assignment approach that allows for assigning different importance to the atoms of a query molecule by weighting the edges of the optimal assignment. Then, we show that particle swarm optimization is able to optimize these edge weights for optimal virtual screening performance. We compared the optimal assignment with optimized edge weights to the original version with equal weights on various benchmark data sets using sophisticated virtual screening performance metrics. The results show that the optimal assignment with optimized edge weights achieved a considerably better performance. Thus, the proposed extension in combination with particle swarm optimization is a valuable approach for ligand-based virtual screening experiments.