Comparison of pKa predictors on in-house dataset
The success of the virtual screening is fundamentally influenced by the optimal tautomer form and ionization state that underlines the importance of accurate pKa prediction for even large libraries. Moreover, ADME and PK optimization supported by in silico methods also requires information on the ionization state, therefore, fast and accurate prediction of pKa values is in high demand. Testing commercially available in silico tools using unbiased compound set provides information about overall performance and uncovers typical errors that might need further attention. Our test set involved ~ 200 experimental pKa values of ~100 compounds that were predicted by different pKa predictors including Marvin, ACD, Epik, Pallas and Pharma Algorithm (ADME Box). Statistical analysis and specific examples will be discussed evaluating the predictive power of different commercially available pKa predictors.