Table 4. Eight consistently well-performing prediction methods based on consistent ranking within the Top 20 according to various statistical metrics.

Submissions were ranked according to RMSE, MAE, R², and τ. Many top methods were found to be statistically indistinguishable considering uncertainties of error metrics. Moreover, sorting of methods was influenced significantly by the choice of metric chosen. We assessed top 20 methods according the each metric to determine which methods are always among the top 20 according to all four statistical metrics chosen. A set of consistently well-performing methods were determined: Four QM-based and four empirical methods. Seven of these methods are blind submissions of SAMPL6 Challenge, and one of them (REF13) is a non-blind reference calculation. Performance statistics are provided as mean and 95% confidence intervals.

ID	Method Name	Category	Type	RMSE	MAE	R2	Kendall’s Tau (t)
hmz0n	cosmotherm_FINE19	Physical (QM)	Blind	0.38 [0.23, 0.55]	0.31 [0.19, 0.46]	0.77 [0.36, 0.94]	0.64 [0.17, 1.00]
gmoq5	Global XGBoost-Based QSPR LogP Predictor	Empirical	Blind	0.39 [0.28, 0.49]	0.34 [0.23, 0.46]	0.74 [0.40, 0.92]	0.59 [0.12, 0.89]
j8nwc	EC_RISM_wet_P1w+2o	Physical (QM)	Blind	0.47 [0.17,0.75]	0.31 [0.15, 0.54]	0.74 [0.33, 0.97]	0.81 [0.46, 1.00]
hdpuj	RayLogP-II, a cheminformatic QSPR model predicting the octanol/water partition coeZcient	Empirical	Blind	0.49 [0.37, 0.61]	0.44 [0.32, 0.57]	0.74 [0.40, 0.94]	0.67 [0.22, 1.00]
dqxk4	LogP_SMD_Solvation_DFT	Physical (QM)	Blind	0.49 [0.33, 0.62]	0.42 [0.26, 0.57]	0.69 [0.35, 0.91]	0.67 [0.27, 0.96]
vzgyt	rfs-logp	Empirical	Blind	0.50 [0.27,0.68]	0.38 [0.21, 0.58]	0.72 [0.29, 0.95]	0.64 [0.23, 0.92]
qyzjx	EC_RISM_dry_P1w+2o	Physical (QM)	Blind	0.54 [0.34, 0.75]	0.46 [0.31, 0.64]	0.73 [0.31, 0.97]	0.78 [0.44, 1.00]
REF13	SlogP (MOE)	Empirical	Reference	0.55 [0.38, 0.71]	0.47 [0.31,0.65]	0.69 [0.29, 0.92]	0.60 [0.08, 0.96]