A new protocol to improve the predictive power of molecular docking

Abstract

Quantitative understanding of molecular recognition is key for basic research and computer-aided drug design projects. Docking mimics ligand-receptor association in silico, providing an atomic-level model structure of their complex. Unfortunately most docking algorithms underestimate receptor flexibility, reducing the rate of success when binding induces large structural changes of partners. Ensembledocking, where a set of receptor structures (e.g. from MD simulations) is considered, was implemented to overcome this limitation. Clearly, ensemble structures should include conformations prone to host ligands (holo form), which is usually not the case when apo and holo forms are separated by high free energy barriers. To improve generation of holo-like receptor conformations starting only from its apo form, we implemented a computational protocol based on enhancedsampling MD simulations. We validated our method on proteins whose apo and holo structures were available and previous efforts to generate holo-like structures and native-like docking poses failed. Receptor structures obtained with our method were comparable to those extracted from MD trajectories of the complex. Furthermore, the docking poses generated by using these structures were native-like and topranked in score.