Table 3.

The performance of NHMC and competitive methods in predicting gene function for different datasets and PPI networks

	All genes
Network		DIP				BioGRID
Method	CLUS-HMC	NHMC		FF	H	NHMC		FF	H
Dataset		α = 0	α = 0.5			α = 0	α = 0.5
seq	0.030	0.025	0.025	0.003	0.002	0.022	0.022	0.004	0.006
pheno	0.021	0.018	0.019	0.002	0.001	0.018	0.018	0.004	0.002
struc	0.018	0.012	0.016	0.002	0.000	0.012	0.012	0.004	0.002
homo	0.040	0.013	0.031	0.001	0.001	0.013	0.013	0.002	0.002
cellcycle	0.017	0.297	0.273	0.004	0.002	0.013	0.013	0.006	0.006
church	0.017	0.013	0.012	0.003	0.002	0.012	0.012	0.006	0.006
derisi	0.018	0.022	0.021	0.004	0.002	0.039	0.315	0.006	0.006
eisen	0.025	0.020	0.020	0.004	0.002	0.021	0.335	0.006	0.008
gasch1	0.020	0.017	0.017	0.003	0.002	0.029	0.339	0.006	0.006
gasch2	0.019	0.020	0.018	0.004	0.002	0.015	0.016	0.006	0.006
spo	0.018	0.019	0.018	0.004	0.002	0.017	0.017	0.006	0.006
exp	0.020	0.017	0.017	0.002	0.002	0.018	0.018	0.006	0.006
Average:	0.022	0.041	0.041	0.003	0.002	0.019	0.094	0.005	0.005

We use the 3-fold cross-validation evaluation schema. The average$\overline{\mathit{\text{AUPRC}}}$ (estimated by 3-fold CV) of the CLUS-HMC (α = 1), NHMC (α = 0.5 and α = 0), FunctionalFlow (FF), and Hopfield (H) methods, when predicting gene function in yeast using GO annotations. We use 12 yeast (Saccharomyces cerevisiae) datasets. Results for two PPI networks (DIP and BioGRID) are presented.