A meta-learning approach for B-cell conformational epitope prediction

BMC Bioinformatics

Table 6 Results of the independent test data

Classifier	TPR	FPR	Precision	Accuracy	F-score	MCC	AUC
SEPPA 2.0	0.289	0.050	0.204	0.922	0.239	0.202	0.765
DiscoTope 2.0	0.930	0.763	0.051	0.266	0.097	0.080	0.699
Bpredictor	0.010	0.007	0.057	0.951	0.017	0.006	0.683
ElliPro	0.826	0.535	0.064	0.480	0.119	0.118	0.696
AAP	0.846	0.641	0.055	0.379	0.104	0.086	0.609
ABCpred	0.507	0.480	0.045	0.519	0.082	0.011	0.530
BCPREDS	0.990	0.874	0.048	0.163	0.091	0.072	0.570
BepiPred	0.761	0.499	0.063	0.512	0.117	0.106	0.656
CBTOPE^a	0.159	0.003	0.681	0.961	0.258	0.317	0.681
LBtope^b	0.632	0.578	0.046	0.431	0.086	0.022	0.575
EPMeta^c	0.129	0.043	0.118	0.922	0.124	0.083	0.595
3-level Stacking	0.194	0.008	0.520	0.958	0.283	0.300	0.793
Cascade	0.199	0.008	0.519	0.958	0.288	0.304	0.789

^aWe selected the parameter value (0.7) of the best-performing CBTOPE on the training data set of 94 antigens, and used the value in the independent test. CBTOPE's performances were markedly lower for ACC, F-score, and MCC (0.708, 0.143, and 0.126), using the default value (-0.3).
^bWe selected the parameter value (42.4) of the best-performing LBtope on the training data set of 94 antigens, and used the value in the independent test. By contrast, LBtope's performances were markedly higher for ACC, F-score, and MCC (0.803, 0.123, and 0.077), using the default value (60).
^cWe selected the parameter value (86) of the best-performing EPMeta on the training data set of 94 antigens, and used the value in the independent test. EPMeta did not provide the default parameter value.

ISSN: 1471-2105