Table 2.

Overview of the performance of the final multitask ensemble model (12_Full), used for the challenge, the singletask ensemble model (11_Full), and several commercial logP prediction tools on the SAMPL7, SAMPL6 and Martel et al. data sets [4]

Method	Dataset	R2	RMSE	Spearman ρ
AlogP	SAMPL7	− 0.30 [− 1.78,0.34]	0.82 [0.59,1.01]	0.42 [− 0.09,0.73]
XlogP3	SAMPL7	0.01 [− 1.12,0.46]	0.72 [0.55,0.87]	0.52 [0.07,0.78]
S+ logP	SAMPL7	0.06 [− 1.23,0.64]	0.70 [0.41,0.93]	0.62 [0.19,0.87]
Model 11_Full	SAMPL7	− 0.17 [− 1.49,0.38]	0.78 [0.52,1.01]	0.60 [0.13,0.86]
Model 12_Full	SAMPL7	0.17 [− 0.95,0.65]	0.66 [0.40,0.89]	0.63 [0.20,0.91]
AlogP	SAMPL6	0.56 [− 0.73,0.84]	0.44 [0.25,0.62]	0.83 [0.32,0.97]
XlogP3	SAMPL6	0.54 [− 0.69,0.82]	0.45 [0.29,0.58]	0.71 [0.05,0.94]
S+ logP	SAMPL6	0.42 [− 1.17,0.80]	0.51 [0.32,0.65]	0.71 [0.03,0.94]
Model 11_Full	SAMPL6	0.71 [− 0.25,0.90]	0.36 [0.24,0.46]	0.85 [0.40,0.99]
Model 12_Full	SAMPL6	0.75 [− 0.08,0.93]	0.34 [0.17,0.46]	0.82 [0.30,0.99]
AlogP	Martel et al.	− 0.15 [− 0.34,− 0.00]	1.27 [1.19,1.34]	0.73 [0.69,0.76]
XlogP3	Martel et al.	0.04 [− 0.11,0.16]	1.16 [1.10,1.21]	0.78 [0.75,0.81]
S+ logP	Martel et al.	− 0.26 [− 0.45,− 0.10]	1.33 [1.26,1.39]	0.71 [0.67,0.75]
Model 11_Full	Martel et al.	− 0.33 [− 0.51,− 0.18]	1.36 [1.31,1.41]	0.74 [0.70,0.77]
Model 12_Full	Martel et al.	− 0.00 [− 0.14,0.12]	1.18 [1.13,1.23]	0.76 [0.73,0.80]

The 95% confidence interval for the different performance metrics is shown between square brackets.