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

BMC Bioinformatics

Table 1 Results for multi-class superfamily prediction in the remote homology detection set-up. Results for the adaptive codes method are reported for a SCOP benchmark data set (67 folds, 74 superfamilies, 544 families, with 802 test sequences).

Method (and optimization target)	Error	Balanced Error	Top 5 Error	Balanced Top 5 Error	Detection Rate at fdr = 1%	Detection Rate at fdr = 5%
PSI-BLAST	0.3990	0.4571	0.2731	0.3654	0.3229	0.4214
one-vs-all: Sfams	0.2706	0.4454	0.1047	0.1973	0.4239	0.5549
one-vs-all: Sfams, Fams	0.2706	0.4454	0.1097	0.2070	0.4239	0.5549
Sigmoid Fitting: Sfams	0.3678	0.5561	0.2020	0.3724	0.3641	0.4726
Adaptive Codes: Sfams (zero-one)	0.2444	0.3805	0.0960	0.1591	0.4264	0.5711
Adaptive Codes: Sfams (balanced)	0.2481	0.3723	0.1110	0.1634	0.4289	0.5673
Adaptive Codes: Sfams, Fams (zero-one)	0.2369	0.3739	0.0948	0.1561	0.4352	0.5698
Adaptive Codes: Sfams, Fams (balanced)	0.2394	0.3632	0.1047	0.1558	0.4302	0.5698
Adaptive Codes: Sfams, Fams, Fams (zero-one)	0.2219	0.3401	0.0910	0.1359	0.4277	0.5723
Adaptive Codes: Sfams, Fams, Fams (balanced)	0.2195	0.3273	0.1047	0.1516	0.4327	0.5910

ISSN: 1471-2105