Assignment of structural domains in proteins using diffusion kernels on graphs

BMC Bioinformatics

Table 3 Comparison of KluDo’s performance with that of four automatic methods over ASTRAL40 separated by the number of domains

	1-domain		2-domain		3-domain		4-domain		Total
	SCOP	CATH	SCOP	CATH	SCOP	CATH	SCOP	CATH	Total
KluDo (LED, KK)	91.1	92.4	70.3	78.1	44.7	48.2	16.5	27.7	84.0
KluDo (LED, SP)	91.1	93.2	66.9	76.4	51.1	52.1	39.6	41.1	84.7
DomainParser	93.1	94.7	76.4	75.3	76.8	56.9	38.9	35.1	85.5
PDP	85.8	89.3	78.4	74.4	76.4	57.8	58.3	47.0	82.1
DDomain	95.0	96.2	72.1	69.1	69.3	50.7	47.2	43.6	84.7
SWORD	90.1	93.1	74.9	71.6	73.9	53.1	50.7	35.1	83.2

KluDo’s accuracy in the case of using the LED kernel with the two clustering methods is compared against four automated methods over ASTRAL40 based on the OL score (considering an 85% boundary consistency threshold). The results are presented both overall and separated by the number of domains according to both SCOP and CATH. The maximum accuracy in each column is depicted in bold. KK and SP stand for kernel k-means and spectral clustering, respectively

ISSN: 1471-2105