To accurately predict binding affinities between protein and ligands from first principles, it is essential to extensively sample the accessible conformations of the dynamic protein-ligand system. Standard free energy methods based on MD simulations typically require days to compute a converged estimate of the free energy of binding for each ligand, as many pair-wise interactions among protein, ligand and solvent atoms have to be calculated at each MD time step. On the contrary, docking concepts are quite efficient in predicting the binding orientation of ligands in a protein's binding site but neglect the dynamic aspects of the protein-ligand complex resulting in inaccurate predictions of binding affinities.
To combine the efficiency of protein-ligand docking with the accuracy of MD-based free-energy methods, we have extended our method Limoc to estimate binding affinities. Due to the pre-generation of an EPS using Limoc, only the ligand degrees-of-freedom in this EPS need to be considered. This decoupling of protein and ligand sampling allows for an efficient prediction of binding affinities for a large set of chemicals potentially binding to the target protein. When Limoc was applied to docking, the method outperformed standard ensemble docking. Furthermore, structural analysis of the Limoc trajectory revealed that it was able to mimic the trajectories of protein-ligand complex MD trajectories for a diverse set of ligands.