Table 6.

External performance of single models for predicting classification endpoints (vT, nT, EPA, GHS)

	Model	SEN	SPE	MCC	BA	#AD	%AD
nT	BRF	0.830	0.848	0.674	0.839	2100	0.728
	aiQSAR	0.723	0.829	0.556	0.776	2567	0.890
	SARpy	0.772	0.724	0.492	0.748	2488	0.863
	GLM	0.779	0.650	0.425	0.714	2884	1.000
vT	BRF	0.856	0.903	0.585	0.880	2103	0.728
	aiQSAR	0.682	0.963	0.619	0.822	2572	0.891
	SARpy	0.710	0.896	0.467	0.803	2613	0.905
EPA	BRF	0.614	0.851	0.405	0.733	2301	0.805
	aiQSAR	0.603	0.857	0.450	0.730	2547	0.891
	HPT-RF	0.616	0.860	0.462	0.738	2180	0.763
GHS	BRF	0.539	0.872	0.342	0.705	1410	0.490
	aiQSAR	0.568	0.895	0.469	0.731	1475	0.512
	HPT-RF	0.569	0.897	0.476	0.733	1291	0.448

For each model, the sensitivity (SEN), the specificity (SPE), the balanced accuracy (BA), the Matthew’s correlation coefficient (MCC), the number (#AD) and the percentage (%AD) of predictions in AD are reported. For multi-category endpoints (EPA and GHS), SEN and SPE are the average of values computed separately for each class, while BA is the arithmetic mean of the average SEN and SPE. The best values for each metric are italicized