. Author manuscript; available in PMC: 2022 Feb 1.

Published in final edited form as: J Comput Aided Mol Des. 2021 Jan 4;35(2):131–166. doi: 10.1007/s10822-020-00362-6

Table 2. Four consistently well-performing prediction methods for macroscopic pK_a prediction based on consistent ranking within the Top 10 according to various statistical metrics.

Submissions were ranked according to RMSE, MAE, R², and τ. Consistently well-performing methods were selected as the ones that rank in the Top 10 in each of these statistical metrics. These methods also have less than 2 unmatched experimental pK_as and less than 7 unmatched predicted pK_as according to Hungarian matching. Performance statistics are provided as mean and 95% confidence intervals.

Submission ID	Method Name	RMSE	MAE	R²	Kendall’s Tau (τ)	Unmatched Exp. pK_a Count	Unmatched Pred. pK_a Count [2, 12]
xvxzd	Full quantum chemical calculation of free energies and fit to experimental pKa	0.68 [0.54, 0.81]	0.58 [0.45, 0.71]	0.94 [0.88, 0.97]	0.82 [0.68, 0.92]	2	4
gyuhx	S+pKa	0.73 [0.55, 0.91]	0.59 [0.44, 0.74]	0.93 [0.88, 0.96]	0.88 [0.8, 0.94]	0	7
xmyhm	ACD/pKa Classic	0.79 [0.52, 1.03]	0.56 [0.38, 0.77]	0.92 [0.85, 0.97]	0.81 [0.68, 0.9]	0	3
8xt50	ReSCoSS conformations // DSD-BLYP-D3 reranking // COSMOtherm pKa	1.07 [0.78, 1.36]	0.81 [0.58, 1.07]	0.91 [0.84, 0.95]	0.80 [0.68, 0.89]	0	0

Table 2. Four consistently well-performing prediction methods for macroscopic pKa prediction based on consistent ranking within the Top 10 according to various statistical metrics.

Table 2. Four consistently well-performing prediction methods for macroscopic pK_a prediction based on consistent ranking within the Top 10 according to various statistical metrics.