Table 5.

Prediction on simulated missing GO annotations under BMA in CC sub-ontology

m		MicroAvgF1	MacroAvgF1	1-RankLoss	Fmax	RAccuracy
1	H →H	96.03 ±0.09	86.84 ±0.19	96.49 ±0.02	95.36 ±0.09	17.12 ±1.89
	M+H →H	96.77 ±0.01	86.86 ±0.23	97.20 ±0.01	95.55 ±0.01	32.55 ±0.16
	M →M	95.48 ±0.04	86.19 ±0.22	93.83 ±0.01	94.98 ±0.04	12.22 ±0.73
	H+M →M	96.57 ±0.06	86.33 ±0.26	97.23 ±0.26	95.15 ±0.06	33.28 ±1.11
2	H →H	89.09 ±0.02	67.85 ±0.37	86.95 ±0.06	87.84 ±0.02	23.04 ±0.17
	M+H →H	90.82 ±0.02	68.47 ±0.37	90.54 ±0.05	88.84 ±0.02	35.24 ±0.13
	M →M	87.31 ±0.06	66.78 ±0.48	82.46 ±0.04	85.95 ±0.06	16.69 ±0.41
	H+M →M	90.53 ±0.01	67.23 ±0.52	91.43 ±0.03	88.16 ±0.01	37.84 ±0.10
3	H →H	82.54 ±0.06	53.68 ±0.25	79.52 ±0.02	81.76 ±0.06	25.08 ±0.27
	M+H →H	85.87 ±0.05	54.78 ±0.27	86.17 ±0.09	83.74 ±0.05	39.38 ±0.20
	M →M	81.45 ±0.05	52.84 ±0.55	76.54 ±0.07	77.69 ±0.05	24.71 ±0.19
	H+M →M	84.02 ±0.10	53.83 ±0.56	82.41 ±0.12	79.60 ±0.10	35.13 ±0.41

q is the number of simulated missing annotations of a gene. H →H directly uses GO annotations of Human to predict annotations of Human genes. M+H →H uses GO annotations of genes from Mouse and Human to predict annotations of Human genes. M →M and H+M →M follow the similar protocol, but make prediction for Mouse genes. The data in boldface is the statistically significant best among these comparing methods for a particular target species, and the significance is checked by paired t-test at 95% confidence intervals