Aggregation strategies of scoring functions for the improvement of the structure-based virtual screening methods performance.

Abstract

Abstract: Individual scoring functions have proved to have fallacies to predict the active compounds because of the complexity in estimating the protein-ligand binding energy. As a result, consensus scoring (CS) has emerged as a promising alternative for the discovery of novel active compounds. During this study, we purpose a new aggregation strategy of six scoring functions (Contact score, Grid-based score, Solvent Accessible Surface Area score, Continuous score and PB/SA score) and evaluate different combinations of them on 102 targets. We used two types of scoring functions for our aggregations: the first used scores with no alteration (raw scores), while the second also used the scores divided by the number of the heavy atoms of the compounds (weighted scores). We evaluated whether out fused scoring functions had better performance than the reference single scoring function. Our results show that our consensus scoring strategies outperform the reference scoring function in 49% of the cases when the raw scores were used and they increased to 61% when the weighted scores were also used. Our CS strategies could be employed for virtual screening studies.