Are QM/MM models systematically improvable? A case study on carbonic anhydrase binding affinities

Hybrid quantum mechanics/molecular mechanics (QM/MM) is the method of choice for modelling large biochemical systems, but their performance depends critically on how they are implemented. This paper examines the sensitivity of QM/MM enzyme-ligand binding energy predictions to key parameters, including QM region size, the incorporation of thermal and solvation effects as well the use of docked poses versus crystal structures. The focus is on carbonic anhydrase which has emerged as an important drug target for a wide variety of pathologies. It is demonstrated that trends in binding affinities are surprisingly insensitive with respect to these parameters. QM/MM geometry optimization of poses obtained from molecular docking enhances predictive utility for yet-to-be synthesized ligands, though outliers arise from structural deviations or non-native interactions. These findings provide practical guidelines for QM/MM applications in carbonic anhydrase and similar metalloenzymes.