Improving pK_a Predictions with Reparameterized Force Fields and Free Energy Calculations

Given the growing interest in designing targeted covalent inhibitors, methods for rapidly and accurately probing pK_as─and, by extension, the reactivities─of target cysteines are highly desirable. Complementary to cysteine, histidine is similarly relevant due to its frequent presence in protein active sites and its unique ability to exist in two tautomeric states. Here, we demonstrate that nonequilibrium free energy calculations can accurately determine the pK_a values of both residues, often outperforming conventional predictors. Importantly, we find that (1) increasing the van der Waals radius of cysteine's sulfur atom, (2) modifying the backbone charges of histidine, and (3) introducing effective polarization by downscaling the side chain partial charges of both residues can all significantly improve pK_a prediction accuracy. Using the modified CHARMM36m force field on the full dataset reduces the prediction error from 2.12 ± 0.27 pK to 1.28 ± 0.15 pK and increases the correlation with experiment from 0.25 ± 0.09 to 0.58 ± 0.08. Similarly, using the modified Amber14SB force field decreases the error from 3.21 ± 0.29 pK to 1.69 ± 0.23 pK and improves the correlation from 0.15 ± 0.10 to 0.36 ± 0.10.