A Unified Framework for pKA Shift Prediction: Establishing a Common Baseline for Molecular Modeling Accuracy

The charge of peptides and proteins, determined by the effective pKA of their residues, is one of the key factors determining their function. Molecular interactions modulate the effective pKAs, shifting them relative to the bare pKA of isolated functional groups. Computational costs of predicting the effective pKAs range from CPU days for all-atom simulations to fractions of a second for the most simplified approaches. Herein, we examined to what extent the higher computational cost is balanced by higher accuracy. As a model system, we used short peptide sequences systematically combining acidic and basic residues, previously studied experimentally and by all-atom simulations. We supplemented the existing results with our own all-atom simulations, a coarse-grained approach, and two fast approximate approaches: FPTS and pepKalc. Our hypothesis that the predictions from all-atom approaches are more accurate than the approximate ones was confirmed only partly. The pKA shifts from most approaches agreed qualitatively but differed quantitatively. Two out of the four all-atom approaches matched the experiments significantly better than the others, while the remaining two were comparable or worse than the approximate ones. Our key finding is that all-atom simulations can be more accurate; however, their accuracy depends not only on the chosen force field, but also on other technical details that are often overlooked. Furthermore, comparing pKA shifts predicted by various simulations requires accounting for different pKA inputs. We show how to shift them to a common reference, enabling for the first time a rigorous and unified comparison across diverse pKA prediction methodologies.