Table 4.

The cross-validation performances of various algorithms using the three-state α-helix (H), β-sheet (E) and other (O) assignment scheme.

Method	SP175						SP175(nr)
	H		E		O		H		E		O
	δ	r	δ	r	δ	r	δ	r	δ	r	δ	r
SELMAT3	0.063	0.957	0.083	0.862	0.078	0.701	0.065	0.954	0.090	0.833	0.083	0.672
SELMAT1_norm	0.062	0.958	0.070	0.904	0.071	0.757	0.065	0.955	0.072	0.897	0.073	0.746
SIMPLS	0.055	0.968	0.070	0.905	0.065	0.800	0.056	0.967	0.071	0.901	0.065	0.797
PCR	0.057	0.966	0.069	0.906	0.066	0.796	0.058	0.965	0.071	0.902	0.066	0.792
PLS	0.053	0.971	0.073	0.895	0.068	0.781	0.053	0.970	0.074	0.893	0.069	0.774
PLS-opt	0.052	0.971	0.070	0.902	0.066	0.796	0.054	0.970	0.072	0.900	0.066	0.790
NN	0.055	0.968	0.067	0.912	0.062	0.816	0.056	0.967	0.068	0.909	0.064	0.805
SIMPL-NN	0.057	0.965	0.064	0.923	0.055	0.860	0.056	0.964	0.065	0.918	0.057	0.850
SVM	0.057	0.966	0.069	0.908	0.066	0.792	0.060	0.964	0.072	0.902	0.067	0.785

The (nr) tag indicates that the cross-validation was carried out under more stringent conditions where no proteins in the training or validation set with the same CATH homologous superfamily as that of the test protein were included. The best results for each secondary structure type are shown in bold.