Theoretical Study of Protein-Ligand Interactions Using the Molecules-in-Molecules Fragmentation-Based Method.

Abstract

We have recently significantly expanded the applicability of our Molecules-in-Molecules (MIM) fragmentation method to large proteins by developing a three-layer model (MIM3) in which an accurate quantum-mechanical method is used in conjunction with a cost-effective, dispersion-corrected semiempirical model to overcome previous computational bottlenecks. In this work, we develop MIM3 as a structure-based drug design tool by application of the methodology for the accurate calculation of protein-ligand interaction energies. A systematic protocol is derived for the determination of the geometries of the protein-ligand complexes and to calculate their accurate interaction energies in the gas phase using MIM3. We also derive a simple and affordable procedure based on implicit solvation models and the ligand solvent-accessible surface area to approximate the ligand desolvation penalty in gas-phase interaction energy calculations. We have carefully assessed how closely such interaction energies, which are based on a single protein-ligand conformation, display correlations with the experimentally determined binding affinities. The performance of MIM3 was evaluated on a total of seven data sets comprising 89 protein-ligand complexes, all with experimentally known binding affinities, using a binding pocket involving a quantum region ranging in size from 250 to 600 atoms. The dispersion-corrected B97-D3BJ density functional, previously known to perform accurately for calculations involving non-covalent interactions, was used as the target level of theory for this work, with dispersion-corrected PM6-D3 as the semiempirical low level to incorporate the long-range interactions. Comparing directly to the experimental binding potencies, we obtain impressive correlations over all seven test sets, with an R2 range of 0.74-0.93 and a Spearman rank correlation coefficient (ρ) range of 0.83-0.93. Our results suggest that protein-ligand interaction energies are useful in predicting binding potency trends and validate the potential of MIM3 as a quantum-chemical structure-based drug design tool.