Table 2. Performance of the standard models by different learning methods on validation sets V^60P+45N and V^60P+60N*.

Data	Model	Sensitivity	Specificity	Accuracy	MCC
V60P+45N	AVPcompo	83.3	88.9	85.7	0.72
	AVPphysico	88.3	82.2	85.7	0.71
	ANNcompo	76.7	75.6	76.2	0.52
	ANNphysico	85.0	80.0	82.9	0.65
	LDAcompo	85.0	44.4	67.6	0.33
	LDAphysico	88.3	53.3	73.3	0.45
	RFcompo	86.7	80.0	83.8	0.67
	RFphysico	93.3	77.8	86.7	0.73
V60P+60N*	AVPcompo#	83.3	98.3	90.8	0.83
	AVPphysico#	93.3	91.7	92.5	0.85
	ANNcompo#	81.7	93.3	87.5	0.76
	ANNphysico#	91.7	90.0	90.8	0.82
	LDAcompo#	78.3	66.7	72.5	0.45
	LDAphysico#	81.7	75.0	78.3	0.57
	RFcompo#	93.3	93.3	93.3	0.87
	RFphysico#	90.0	95.0	92.5	0.85

V60P+60N^* contained non-experimental peptides. The models trained by T544P+544N^* were marked by the number sign #; the models trained by T544P+407N had no marks. All the AVP models here were built using SVM [23].