Table 4.

Comparative analysis of ab initio methods of protein domain boundary prediction

Predictor name and reference	Training data	Domain boundary annotation	Test data and results
PPRODO [7]	Nonredundant set of 522 proteins	Domain boundary is considered to be ± 20 residues from the true domain boundary; these residues are considered to be positive. The other (nonboundary, i.e., domain) residues are considered to be negative.	CASP-5 10-fold cross-validation experiment Aaccuracy of about 66 %
DOMPro [8]	CATH (2.5.0)		CAFASP-4 Specificity 0.71 Sensitivity 0.71
DROP [17]	DS-All	Linker (nondomain) predictor.	CASP8 NDO score: 0.76
Dobo [16]	CATH(3.3.0)	Domain-boundary definitions from PSI-BLAST	CASP9 Multidomain Recall: 0.72 Precision: 0.68 Accuracy: 0.82
Current work: PDP-CON (1-star consensus SVM)	CATH (2.5.1)	True domain definitions obtained from CATH (2.5.1)	Accuracy: CASP8: 0.91 CASP9: 0.90 CASP10: 0.79 (see Table 3 for details)
Current work: MLP-1 (with k = 20)	CATH (2.5.1)	Domain boundary is considered to be ± 20 residues from the domain definition obtained from CATH (2.5.1); domain boundary residues vare considered to be positive and the other (domain) residues are negative	Recall: CASP8: 0.94 CASP9: 0.94 CASP10: 0.96 (see Table 5 for details)