The Effect of Density Functionals and Basis Sets on <scp>QM</scp> ‐Cluster Models of Chorismate Mutase

ABSTRACT This work examined three aspects of computational enzymology: (1) the convergence of kinetics and thermodynamics of quantum mechanical (QM)‐cluster models with respect to model size and composition, (2) the accuracy of predicted values compared to experimental kinetics, and (3) finding a tractable level of electronic structure theory for computational enzymology workflows when a model size of over 200 atoms is necessary. Using 10 QM‐cluster models of the Bacillus subtilis chorismate mutase active site, we seek a quantum chemical level of theory that will improve accuracy of atomic‐level enzymology studies without a significant increase in resource demands. Values of and were calibrated with 31 types of density functional theory (DFT), 5 semi‐empirical methods, and 20 one‐electron basis sets. Computed barrier heights were compared to an experimentally determined of 15.4 ± 0.5 kcal mol −1 at 25°C to assess performance. The results showed a clear trend that hybrid GGA and hybrid meta GGA approximations, notably both M05‐2X and CAM‐B3LYP with mixed 6–31G()/6–31G basis sets, were well‐converged with respect to model size, and within 1 kcal mol −1 of the experimental value. Next, B3LYP computations with larger basis sets increased computed values. Our work suggests an approach to obtain high‐quality kinetic and thermodynamic data for enzyme mechanism studies using current generation density functionals without also having to use computationally demanding basis sets. If computational resources are available, the most accurate levels of theory for chorismate mutase QM‐cluster models are B97‐2 or PW6B95 with 6–311+G(d) basis sets. Otherwise, geometry optimizations with B97‐2 or PW6B95 and mixed 6–31G()/6–31G basis sets followed by single point computations with the selected functional and larger 6–311+G(d) basis sets give reliable kinetics and thermodynamics. The semi‐empirical GFN2‐xTB approach provided impressive speed and accuracy, but gave concerning results when used in composite approaches with more rigorous levels of theory. The recommended levels of theory may not retain their beneficial error cancellation when applied alongside statistical sampling of enzyme conformations and reaction dynamics, thereby warranting further calibration.