KEGG orthology prediction of bacterial proteins using natural language processing

BMC Bioinformatics

Table 2 Performance comparison with other KO annotation tools

Method	Match	Unmatch	Missed	Added	Precision	Recall	F1
BlastKOALA	6172	64	552	100	0.974	0.909	0.941
GhostKOALA	6423	26	339	117	0.978	0.946	0.962
KofamKOALA	5955	88	745	953	0.851	0.877	0.864
Ours	6426	183	179	171	0.948	0.947	0.947
Ours w/o classifier	5943	62	783	407	0.876	0.927	0.900
Ours with threshold	6399	169	220	151	0.952	0.943	0.948

Best performance is marked in bold. We calculated the number of match (predicted KO is identical to the KO defined in KEGG GENES), unmatch (predicted KO is different from the KO defined in KEGG GENES), missed (the KO is defined in KEGG GENES but no prediction was made), and added (no KO is defined in KEGG GENES, but the prediction assigned a K number) for each tool, along with precision, recall, and F1 score

ISSN: 1471-2105