Successes and Pitfalls in Scoring Molecular Interactions

Abstract

The appropriate evaluation of molecular interactions is still an important challenge in molecular modeling field. The difficulties for scoring those interactions become clear when the performance of the so-called ‘ligand-based’ methods is compared with the ‘structure-based’ methods. Although more information is provided for the latter, the former very often performs better in recovering actual binders from a set of ligands and decoys. Here, we compare some results of different scoring functions as implemented and tested in our hybrid-docking algorithm named LiBELa. The results show that the parameters devised from force fields such as AMBER provide a very good estimate of the polar and van der Waals contribution for binding. When properly set up, soft-docking, i.e., a smooth potential, can, at best, reproduce the results obtained with force fields, but hardly outperforms the former. Finally, the results obtained in our group and from other groups clearly indicate that an adequate potential for modeling the solvent effect is still a goal to be achieved. At best, the current empirical solvation models used in docking algorithms can lead to improvement of enrichment in ROC curves for a fraction (50% or less) of the current and gold-standard test sets. In conclusion, the scoring of molecular interactions at an atomic level is a promising field with many important advances achieved but also with a number of open issues.