Predicting Carbonic Anhydrase Binding Affinity: Insights from QM Cluster Models.

Abstract

A systematic series of QM cluster models has been developed to predict the trend in the carbonic anhydrase binding affinity of a structurally diverse dataset of ligands. Reference DLPNO-CCSD(T)/CBS binding energies were generated for a cluster model and used to evaluate the performance of contemporary density functional theory methods, including Grimme's "3c" DFT composite methods (r²SCAN-3c and ωB97X-3c). It is demonstrated that when validated QM methods are used, the predictive power of the cluster models improves systematically with the size of the cluster models. This provided valuable insights into the key interactions that need to be modeled quantum mechanically and could inform how the QM region should be defined in hybrid quantum mechanics/molecular mechanics (QM/MM) models. The use of r²SCAN-3c on the largest cluster model composed of 16 residues appears to be an economical approach to predicting binding trends compared with using more robust DFT methods such as ωB97M-V and provides a significant improvement compared with docking.