Protein-RNA interface residue prediction using machine learning: an assessment of the state of the art

BMC Bioinformatics

Table 8 Protein-based evaluation of Structure Methods on Structure Data

	IDStr	IDStr	IDStr	PSSM Str	PSSM Str	PSSM Str	Smo PSSM	Smo PSSM	Smo PSSM
	NB	LK	RBFK	NB	LK	RBFK	Str NB	Str LK	Str RBFK
RB106Str	0.72 (3)	0.71 (7)	0.72 (3)	0.71 (7)	0.72 (3)	0.72 (3)	0.69 (9)	0.71 (7)	0.72 (3)
RB144Str	0.71 (6.5)	0.71 (6.5)	0.71 (6.5)	0.71 (6.5)	0.72 (3.5)	0.73 (2)	0.68 (9)	0.72 (3.5)	0.74 (1)
RB198Str	0.72 (3.5)	0.72 (3.5)	0.72 (3.5)	0.68 (8)	0.72 (3.5)	0.72 (3.5)	0.66 (9)	0.71 (7)	0.72 (3.5)
Average	0.72 (4.3)	0.68 (5.7)	0.69 (4.3)	0.70 (7.2)	0.72 (3.3)	0.72 (2.8)	0.68 (9)	0.71 (5.8)	0.73 (2.5)

AUC (averaged over five folds) of structure methods on structure data using residue-based evaluation. Based on average rank, the best structure method across the three datasets is the SVM classifier that uses the RBF kernel and SmoothedPSSMStr (with a window size of 3) as input (NB - Naïve Bayes, SVM - Support Vector Machine, LK - Linear Kernel, RBFK - Radial Basis Function Kernel) The rank of each classifier is shown in parentheses.

ISSN: 1471-2105