Detection of protein catalytic residues at high precision using local network properties

BMC Bioinformatics

Table 1 Comparison of the performance of predictions of catalytic residues using different scoring functions and threshold values on the extended protein set

	F ₁	F ₂	Precision	Coverage
Closeness + RSA^a	15.13%	11.54%	8.22%	31.66%
Eq. 1^b, MDev1	20.82%	-	15.42%	32.05%
Eq. 1, MDev2	-	20.56%	28.10%	9.91%

Calculations were run over the residue interaction networks derived from the 226 protein structures from our extended test set. Scoring functions used here are described in Methods. Values for precision and coverage were obtained over the whole set. F₁ and F₂ respectively represent the F-measure defined in Methods when using β = 1 and β = 2. ^a As proposed by Amitai [14]. RSA: residue surface accessibility. ^bResidues were considered as catalytic if their MDev value for the scoring function defined in Equation 1 was superior to the indicated threshold value. Corresponding 'specificity' (equal to (p_-, r_-)/r_-) values were 97.80% at MDev1, 99.68% at MDev2, and 95.57% when using closeness combined to RSA.

ISSN: 1471-2105