Improving automatic GO annotation with semantic similarity

BMC Bioinformatics

Table 1 Comparison of GrAPFI-GO performance with and without node neighborhood-based similarity

Method	GO aspects	Precision	Recall	F1-score
GrAPFI	BP	0.28	0.21	0.24
	MF	0.23	0.28	0.25
	CC	0.33	0.31	0.32
GrAPFI-CN	BP	0.26	0.20	0.23
	MF	0.26	0.21	0.23
	CC	0.32	0.30	0.31
GrAPFI-JA	BP	0.26	0.19	0.23
	MF	0.21	0.27	0.24
	CC	0.32	0.30	0.31
GrAPFI-PA	BP	0.25	0.19	0.22
	MF	0.21	0.26	0.23
	CC	0.31	0.29	0.30
GrAPFI-SA	BP	0.26	0.20	0.23
	MF	0.21	0.26	0.23
	CC	0.33	0.30	0.31
GrAPFI-SO	BP	0.26	0.20	0.23
	MF	0.21	0.26	0.23
	CC	0.33	0.30	0.31
GrAPFI-HDI	BP	0.26	0.19	0.22
	MF	0.21	0.26	0.23
	CC	0.32	0.30	0.31
GrAPFI-HPI	BP	0.26	0.19	0.22
	MF	0.21	0.26	0.23
	CC	0.33	0.30	0.31
GrAPFI-LLHN	BP	0.28	0.21	0.24
	MF	0.22	0.27	0.24
	CC	0.33	0.31	0.32

GrAPFI-CN: common neighbor, GrAPFI-JA: Jaccard index, GrAPFI-PA: preferential attachment, GrAPFI-SA: Salton index (), GrAPFI-SO: Sorensen index, GrAPFI-HDI: hub depressed index, GrAPFI-HPI: hub promoted index, GrAPFI-LLHN: local Leicht–Holme–Newman index. In each case, we report average precision, recall and F1-score for three GO aspects, namely Biological Process (BP), Molecular Function (MF) and Cellular Component (CC). The experiment is run with cut-off score of 0.5, minimum similarity threshold of 0.3

ISSN: 1471-2105