Using Molecular Dynamics and Enhanced Sampling to Predict Binding Poses Beyond The Rigid-Docking Approximation

Sammanfattning: A computational method is described and tested for prediction of ligand-binding poses between the human farnesoid X receptor and a set of 36 potential agonists, provided by the D3R Grand Challenge 2016. Using tools such as Molecular Docking, Molecular Dynamics, Reconnaissance metadynamics and cluster analysis, the method is an attempt to predict the binding pose without being biased by experimental data. When comparing the predicted poses with the crystal structures, more than half of the ligands were predicted accurately. It is shown that the accuracy of the Molecular Docking is very conformation dependent, as the flexibilty of two α-helices adjacent to the active site makes it difficult for docking to predict the correct pose. Molecular Dynamics are dependent on the predictions from docking, and the force field (GAFF) used for the ligands may be the reason for that only 3 of the accurately predicted poses were refined further. Reconnaissance metadynamics did not result in finding any better poses with the collective variables set used. More effort is needed to determine a better set of collective variables, which are able to take the flexibility of the α-helices and the positions of the side chains into consideration, as well as possibly enable Reconnaissance metadynamics to overcome the short-comings of docking.