Energy Decomposition Analysis of Protein-Ligand Interactions Using Molecules-in-Molecules Fragmentation-Based Method.

Abstract

Accurate prediction of protein-ligand binding affinities and their quantitative decomposition into residue-specific contributions represent challenging problems in drug discovery. While quantum mechanical (QM) methods can provide an accurate description of such interactions, the associated computational cost is normally prohibitive for broad-based applications. Recently, we have shown that QM-based protein-ligand interaction energies in the gas phase can be determined accurately using our multilayer molecules-in-molecules (MIM) fragmentation-based method at a significantly lower computational cost. In this paper, we present a new approach for calculating protein-ligand interactions using our three-layer model (MIM3) that allows us to decompose the total binding affinity into quantitative contributions from individual residues (or backbone and side chain), crystal water molecules, solvation energy, and entropy. In our approach, the desolvation energy and entropy changes during protein-ligand binding are modeled using simple and inexpensive empirical models while intermolecular interactions are computed using an accurate QM method. The performance of our approach has been assessed on a congeneric series of 22 thrombin inhibitors, all with experimentally known binding affinities, using a binding pocket cutout of 120 residues with more than 1550 atoms. Comparison of our MIM3-calculated binding affinities calculated at the B97-D3BJ/6-311++G(2d,2p) level with experiment shows a good correlation with an R2 range of 0.81-0.88 and a Spearman rank correlation coefficient (ρ) range of 0.84-0.89 while providing a quantitative description of residue-specific interactions. We show that such residue-specific interaction energies can be employed to identify and rationalize both obvious (e.g., hydrogen bonds, π···π) and nonobvious (e.g., CH···π) interactions that play a critical role in protein-ligand binding. We suggest that such quantitative information can be used to identify the key residues that determine the comparative binding affinities of different ligands in order to improve and optimize the effectiveness of computational drug design.