Can docking scoring functions guarantee success in virtual screening?

Abstract

Computational methods such as virtual screening have an important role in drug discovery. Virtual screening is frequently used for the identification of hit compounds at early stages of the drug discovery pipeline. In this sense, molecular docking is one of the most extended tools for performing virtual screening, with scoring functions being a critical component of this tool. Part of the success of docking-based virtual screening campaigns can be attributed to the performance of scoring functions. Scoring functions are applied for the prediction of affinities in receptor-ligand complexes, the ranking of compounds according to binding energies and the discrimination between active and inactive chemicals. Because of the complexity of molecular recognition processes, large approximations are required in scoring functions to achieve their predictions in a reasonable time. These approximations lead to unavoidable inaccuracies in scoring functions that negatively affect their performance. Here are presented the main foundations behind current scoring functions and their limitations. The results of comparative studies between scoring functions are discussed as well as strategies for overcoming the inaccuracies and limitations associated with the use of scoring functions. Finally, we provide some ideas that could be useful for selecting a scoring scheme for structure-based virtual screening and a brief outlook on the expected future developments that will improve the performance of current scoring functions.