Computational prediction of receptor-ligand binding affinity in drug discovery

Abstract: The evaluation of inhibition constants or, more generally, receptor-ligand binding affinities is a crucial part of the drug discovery process. Chemical synthesis and affinity screening is only affordable for a limited number of compounds. This makes computational methods to predict binding affinities of candidate ligands highly desirable.A linear interaction energy (LIE) method, developed earlier for prediction ofabsolute binding free energies from potential energy averages collected frommolecular dynamics (MD) simulations is used to calculate the affinity of a set ofmethotrexate analogues for human dihydrofolate reductase (DHFR) and for twomutants of this enzyme. DHFR is a target of drug intervention in the therapy of, forexample, cancer and rheumatoid arthritis as well as microbial infections, and thereforealso the focus of intense ligand design efforts. The LIE method is examined closelyand an improved method is constructed by considering systematic deviations fromelectrostatic linear response for polar neutral compounds.To facilitate further free energy calculations, a new MD program is developed. It is written for efficient calculations in biomolecular systems using the free energy perturbation (FEP), empirical valence bond and LIE methods. Using the new software, FEP simulations are carried out to predict relative free energies of binding for a series of candidate DHFR inhibitors and also to give insight into the structural features affecting binding. The potencies of two compounds selected and synthesised are in fair agreement with predictions. Another ligand design project employs LIE calculations on inhibitors of thrombin (a target of drug therapy of blood coagulation disorders). With the addition of a constant term in the LIE equation, absolute binding free energies are well reproduced.As an attempt to find more efficient yet reliable methods for binding affinityprediction, a combination of empirical scoring and conformational sampling is introduced and tested on three different protein-ligand complexes.