• pKa;
  • aqueous solution;
  • free energy;
  • drug molecules;
  • conformational ensemble


We have investigated different protocols for calculation of pKa values for drug-like organic molecules, using quantum mechanical methods. In particular, we examined the role of geometry optimization schemes and the use of empirical correction terms. A general parameterization, common for all compound classes, gave a mean unsigned error of 0.6 pKa units when applied to a test set containing 45 drug molecules. We also investigated the accuracy dependence of separate functional groups with respect to different geometry optimization methods. While heterocyclic compounds showed a consistent improvement with increasing theoretical rigor, other groups, such as amines or carboxylic acids, did not. We also investigated models using local atomic properties, rather than the free energy of solvation of the whole solute, by creating one-component regression models using simple parameters calculated from the neutral ground state in vacuum. For some functional groups, these regression models showed similar performance as explicit titration calculations using continuum solvent models and explicit treatment of the ionized state. This finding indicates that the limiting factors for many protocols used to calculate pKa shifts may be dominated by noise from the conformational ensemble and inadequate description of the ionic solvated state. © 2005 Wiley Periodicals, Inc. Int J Quantum Chem, 2005