Virtual Screening of Drugs: Score Functions, Docking, and Drug Design

The computational approach for new drug design and/or identification, was initially proposed in mid 70's. The virtual screening of chemical libraries against a biological target has proven its reliability on structure-based drug design, for instance, for many HIV virus protein inhibitors and for the development of Cyclin-Dependent Kinase inhibitors. Target- based virtual screening, allied to docking studies, enables searches on larger data set of probable ligands, with less costs than the traditional experimental screening. The increasing availability of small molecules databases and its free online distribution is now allowing not only pharmaceutical industries, but independent research labs as well, to apply this methodology on early stages of drug discovery. When the protein target structure is available, and a chemical virtual library is accessible, following questions need to be answered: how the target and the ligand interact and how these interactions may be evaluated? Several docking algorithms for the identification of the molecular features responsible for binding specificity are available. While such algorithms are very robust and accurate, the scoring functions remain more questionable in the sense of what parameters should be considered when defining protein-ligand binding affinity when ranking candidates pointed-out by the virtual screening to the next step on drug testing. Aside conformational and chemical information, pharmacokinetics properties should be considered as well when selecting potential new drugs. Along with structural well-match, appropriate molecular features that define desired kinetics characteristics should be consistently addressed for usefulness of virtual screening results. The present review is focused on these questions and their implication for virtual screening.