Table 3.

Results of cross-validating the different SVMs by loo

Specificity of SVM	Positive training points	Kernel type	Leave-one-out cross-validation				Quality of SVM
			Error	Sn	Sp	MCC
Large clusters	282 Labeled and 664 unlabeled data points (18 + 646)
Dhb=Sal	11	l	0.4	100	92	96	++
Asp=Asn=Glu=Gln=Aad	43	r	1.4	100	91	95	++
Pro=Pip	20	r	0.7	90	100	95	++
Cys	17	r	0.7	100	89	94	++
Ser=Thr=Dhpg=Dpg=Hpg	50	r	2.5	96	91	92	++
Gly=Ala=Val=Leu=Ile=Abu=Iva	92	r	4.3	95	93	90	+
Orn=Lys=Arg	16	l	0.7	88	88	87	+
Phe=Trp=Phg=Tyr=Bht	33	r	3.2	88	85	85	0
Small clusters	273 Labeled and 673 unlabeled data points (27 + 646)
Dhb=Sal	11	l	0	100	100	100	++
Aad	7	l	0	100	100	100	++
Glu=Gln	15	l	0	100	100	100	++
Dhpg=Dpg=Hpg	20	l	0.4	100	95	97	++
Ser	13	l	0.4	92	100	96	++
Cys	17	l	0.7	100	89	94	++
Thr	16	l	0.7	94	94	93	++
Pro	16	r	0.7	94	94	93	++
Asp=Asn	21	l	1.1	90	95	92	++
Val=Leu=Ile=Abu=Iva	60	l	2.9	92	95	91	+
Orn	8	l	0.7	88	88	87	+
Gly=Ala	32	l	3.3	81	90	84	0
Tyr	18	r	2.2	94	77	84	0
Arg	5	l	0.7	80	80	80	0
Phe=Trp	14	l	3.7	57	67	60	0

The more training data that are available the more reliable the trained predictive models are. The ‘quality of SVM’ in the last column, therefore, is a qualitative measure for the MCC. Kernel type l stands for linear kernel and r stands for radial basis function kernel. Error rate, sensitivity (S_n), specificity (S_p) and Mathews correlation coefficient (MCC) are given in percentage.