SVM-Fold: a tool for discriminative multi-class protein fold and superfamily recognition

BMC Bioinformatics

Table 3 Results for multi-class fold prediction in the fold recognition set-up. Results for the adaptive codes method are reported on a SCOP benchmark data set (26 folds, 303 superfamilies, 614 test sequences).

Method (and optimization target)	Error	Balanced Error	Top 5 Error	Balanced Top 5 Error	Recognition Rate at fdr = 1%	Recognition Rate at fdr = 5%
PSI-BLAST	0.6482	0.7029	0.5179	0.5431	0.0814	0.0961
one-vs-all: Folds	0.4625	0.6282	0.1450	0.2345	0.1368	0.2704
one-vs-all: Folds, Sfams	0.4625	0.6282	0.1450	0.2345	0.1368	0.2704
Sigmoid Fitting: Folds	0.4446	0.6103	0.1547	0.2960	0.1336	0.2166
Adaptive Codes: Folds (zero-one)	0.4023	0.5556	0.1059	0.1543	0.1906	0.2655
Adaptive Codes: Folds (balanced)	0.3664	0.5158	0.1075	0.1387	0.1612	0.2785
Adaptive Codes: Folds, Sfams (zero-one)	0.4104	0.5525	0.1107	0.1719	0.2003	0.2329
Adaptive Codes: Folds, Sfams (balanced)	0.3616	0.5153	0.1010	0.1263	0.2068	0.2508
Adaptive Codes: Folds, Sfams, Fams (zero-one)	0.4007	0.5427	0.1075	0.1788	0.2068	0.2557
Adaptive Codes: Folds, Sfams, Fams (balanced)	0.3648	0.5000	0.1091	0.1453	0.2134	0.3013

ISSN: 1471-2105