Table 3.

Statistical quality for models of HO-1 pIC₅₀.

Model	Split	Set	T*	N*	n	r2	q2	s	Fcalc	F(0.05,1,n–2)	p-value
Hybrid	Split 1	Sub-training	0	41	131	0.8085	0.8033	0.337	545	253.33	0.034
		Calibration			131	0.8029	0.7971	0.390	526	253.33	0.035
		Test			60	0.8183	0.8053	0.381	261	252.12	0.049
		Validation			60	0.8291		0.398	281	252.12	0.047
	Split 2	Sub-training	0	33	131	0.7782	0.7721	0.414	453	253.33	0.037
		Calibration			131	0.8187	0.8130	0.349	582	253.33	0.033
		Test			60	0.8888	0.8801	0.302	464	252.12	0.037
		Validation			60	0.7940		0.515	223	252.12	0.053
	Split 3	Sub-training	2	30	131	0.7263	0.7177	0.427	342	253.33	0.043
		Calibration			131	0.7265	0.7192	0.438	343	253.33	0.043
		Test			60	0.8189	0.8037	0.502	262	252.12	0.049
		Validation			60	0.8204	0.562		265	252.12	0.049
SMILES	Split 1	Sub-training			131	0.6933	0.6849	0.427	292	253.33	0.047
		Calibration			131	0.6590	0.6497	0.505	249	253.33	0.050
		Test			60	0.5008	0.4714	0.569	58	252.12	0.100
		Validation			60	0.6290		0.551	98	252.12	0.080
	Split 2	Sub-training			131	0.6508	0.6415	0.520	240	253.33	0.051
		Calibration			131	0.6883	0.6790	0.455	285	253.33	0.047
		Test			60	0.7593	0.7400	0.446	183	252.12	0.059
		Validation			60	0.4645		0.688	50	252.12	0.112
	Split 3	Sub-training			131	0.6141	0.6010	0.507	205	253.33	0.057
		Calibration			131	0.6096	0.5994	0.527	201	253.33	0.056
		Test			60	0.6697	0.6431	0.620	118	252.12	0.073
		Validation			60	0.5006		0.614	58	252.12	0.104
Graph	Split 1	Sub-training	0	40	131	0.7115	0.7035	0.414	318	253.33	0.047
		Calibration			131	0.7077	0.6980	0.470	312	253.33	0.045
		Test			60	0.6839	0.6616	0.483	126	252.12	0.071
		Validation			60	0.6751		0.502	121	252.12	0.072
	Split 2	Sub-training	0	34	131	0.6717	0.6613	0.504	264	253.33	0.049
		Calibration			131	0.7293	0.7209	0.444	348	253.33	0.043
		Test			60	0.7247	0.6941	0.452	153	252.12	0.064
		Validation			60	0.7021		0.543	137	252.12	0.068
	Split 3	Sub-training	2	70	131	0.7336	0.7247	0.421	355	253.33	0.042
		Calibration			131	0.7336	0.7263	0.441	355	253.33	0.042
		Test			60	0.7070	0.6811	0.573	140	252.12	0.067
		Validation			60	0.5712		0.659	77	252.12	0.090

T^* and N^* are preferable values for the threshold and the number of epochs, respectively; n is the number of compounds in the set; r² is the correlation coefficient; q² is the cross-validated correlation coefficient; s is the root-mean-square error; F is the Fisher F ratio; F_{(0.05,1,n–2)} is the 0.05-quantile of the Fisher's distribution F_(1,n–2); p-value is the Fisher test's significance level.