Comparative Evaluation of in Silico pK_a Prediction Tools on the Gold Standard Dataset

Abstract

The predictive performance of five different pK a prediction tools (ACDpKa, Epik, Marvin pKa, Pallas pKa, and VCCpKa) was investigated on the 248-membered Gold Standard dataset. We found VCC as the most predictive, high throughput pK a predictor. However since VCC calculates pK a for the most acidic or basic group only we concluded that ACD and Marvin are in fact the method of choice for medicinal chemistry applications. Analyzing the common outliers we identified guanidines, enolic hydroxyl groups and weak acidic NHs as most problematic moieties from prediction point of view. Our results obtained on the high quality, homogenous Gold Standard dataset could be useful for end-users selecting a suitable solution for pK a prediction.

Comparative evaluation of pKa prediction tools on a drug discovery dataset

“…It is worth noting that this performance is significantly lower than that obtained on public datasets [12,[15][16][17][18]. These results, typically less than half of the cases predicted within 0.5 pK a unit, calls for further development in this field.…”

The Thermodynamic Overlapping pKa of the Antitumor Drug Ibrutinib Using Multiwavelength UV/VIS-Spectroscopy and Potentiometry

“…All determined dissociation constants are not in a good agreement with the predicted values from program MARVIN [25,27] as stated in the results and conclusion chapter. This discrepancy might be caused by unclear resonance structure of the heterocyclic molecule core, and, consequently, different electrones distribution, which can further lead to different predicted values according to the proposed structure.…”

“…Moreover, these values can differ from the experimentally determined dissociation constants. In such cases prediction programs MARVIN [25,27] and ACD/pK [24][25][26] may fail and experimental laboratory detemination is needed. As both potentiometric and spectrophotometric results are similar and regarding the goodness of fit tests one can conclude the obtained experimental results are reliable and they show the real dissociation of the substance.…”

“…Nine commercially available or free programs for predicting ionization constants were compared [4]. Meloun et al [22] used the REGDIA regression diagnostics algorithm written in S-Plus [23] to critically examine the accuracy of pK a predictions with four programs (ACD/pK [24][25][26], Marvin [25,27], PALLAS [25,28] and SPARC [29,30]) considered the best. Balogh et al [25] found the most predictive and reliable predictors to be MARVIN and ACD/Percepta [26].…”

Role of Deprotonation Free Energies in pKa Prediction and Molecule Ranking

“…60 The proposed approach is extremely fast (a set of 10 molecules can be easily studied in 1 h on a typical high-end PC) yet is rigorously founded on the relevant thermodynamic cycles. Another advantage of this predictive method is that the simulations can in principle be performed in the same way for any solvent or a mixture of solvents where the dielectric constant is known, but the experimental pK a data may be scarce.…”