Identification of novel DNA repair proteins via primary sequence, secondary structure, and homology

BMC Bioinformatics

Table 3 Pairwise comparison of classifiers.

Methodology	Test	P	PS	PH	PSH
PS	T	0.7153
	W	0.8413
PH	T	0.0844	0.1287
	W	0.0952	0.1508
PSH	T	0.1006	0.1705	0.7109
	W	0.1508	0.2222	0.8413
BLAST	T	0.0094	0.0104	0.2442	0.0193
	W	0.0159	0.0159	0.0952	0.0556

Using both the (parametric) t-test (T) and the (non-parametric) Wilcoxon Signed-Rank Test (W) on the original GO-PDB dataset, the probability that each pair of classifiers produces the same AUC score on average is listed. The results indicate that the various SVM formulations are not significantly different statistically, though there is a statistical contrast to BLAST (indicated in bold), and SVM formulations including BLAST result in lower probabilities of producing results identical to non-homology feature vector formulations. Bold values indicate the statistical difference is significant at the 0.05 confidence level and we can reject the null hypothesis that the classifiers are equivalent on average.

ISSN: 1471-2105