Improving the efficiency of virtual screening by filtering ligands conformations through a new shape-matching algorithm.

Modern drug design relies crucially on the accurate knowledge of the structures formed by a putative drug with its receptor(s). Computational methods such as molecular docking, which aims to reproduce drug-receptor complexes in silico, have become established tools in this context. Unfortunately, the accuracy of docking is severely constrained by conformational changes taking place upon binding, which are rarely considered by standard methods. Recently, we proposed "EDES", a new computational protocol able to generate bound-like conformations of several protein targets. Accurate description of ligand flexibility is equally crucial, particularly in a virtual screening (VS) context where structures of thousands of ligands are generated without considering any structural adaptation functional to binding. To address this issue, we developed a strategy to select, from a set of ligand structures, only those that snugly fit into the binding site of a receptor. We tested the method on the DUD-e dataset, widely employed to benchmark VS pipelines. Coupled with EDES, our method predicted the true complex structure of all the systems tested while saving a large fraction of computational time.