Table 2.

Statistics Results of Modeling Methods Compared to Previous Model Results

	training set			test set			validate set				i-val set
model	r 2a	MAEb	SEc	r 2	MAE	SE	r 2	MAE	SE	±2SE	% TPe
previous ANNf	0.98	25.3	32.6	0.87	51.4	65.6	0.91	42.7	54.8	219	78% g
ANN-t test optimizedh	0.97	27.7	36.1	0.95	38.8	50.0	0.94	42.1	54.1	216	93%
ANN-v validate optimizedi	0.98	29.8	38.6	0.94	47.0	60.6	0.95	39.3	50.7	203	94%

^aSquare of the Pearson correlation coefficient.

^bMean absolute error.

^cStandard error.

^d±2 standard error filter range from validate SE.

^ePercent true positive predictions of independent validate set.

^fResults from previous study¹⁹ based on 390 compounds.

^g1492 compound i-val set predicted by the previous model.

^hANN model with test set stopping and selection.

ⁱANN model with validate set stopping and selection.