posted on 2019-08-15, 19:36authored byJuan Pablo Arcon, Lucas A. Defelipe, Elias D. Lopez, Osvaldo Burastero, Carlos P. Modenutti, Xavier Barril, Marcelo A. Marti, Adrian G. Turjanski
Virtual screening of large compound
databases, looking for potential
ligands of a target protein, is a major tool in computer-aided drug
discovery. Throughout the years, different techniques such as similarity
searching, pharmacophore matching, or molecular docking have been
applied with the aim of finding hit compounds showing appreciable
affinity. Molecular dynamics simulations in mixed solvents have been
shown to identify hot spots relevant for protein–drug interaction,
and implementations based on this knowledge were developed to improve
pharmacophore matching of small molecules, binding free-energy estimations,
and docking performance in terms of pose prediction. Here, we proved
in a retrospective manner that cosolvent-derived pharmacophores from
molecular dynamics (solvent sites) improve the performance of docking-based
virtual screening campaigns. We applied a biased docking scheme based
on solvent sites to nine relevant target proteins that have a set
of known ligands or actives and compounds that are, presumably, nonbinders
(decoys). Our results show improvement in virtual screening performance
compared to traditional docking programs both at a global level, with
up to 35% increase in areas under the receiver operating characteristic
curve, and in early stages, with up to a 7-fold increase in enrichment
factors at 1%. However, the improvement in pose prediction of actives
was less profound. The presented application makes use of the AutoDock
Bias method and is the only cosolvent-derived pharmacophore technique
that employs its knowledge both in the ligand conformational search
algorithm and the final affinity scoring for virtual screening purposes.