Probabilistic Inference of Biological Networks via Data Integration

Table 2

Cross-validation results for the pairwise kernels using unweighted (U) and weighted (W) combinations of the six unpaired kernels for data sets of different sizes. Shown is test accuracy averaged over , , or data sets (1,098, 2,196, or 4,392 examples, respectively, split into 80% training and 20% test sets). In many cases, the MKL weights yield a significant improvement while in other cases there is no significant change. Significant values are denoted as follows: Wilcoxon signed rank or , and paired -test . Statistically significant values are marked in bold type.

Kernel
	U	W	U	W	U	W
	0.826		0.860	0.867	0.895	0.901
	0.667	0.662	0.663		0.694
	0.764		0.802		0.852
	0.731	0.740	0.756	0.764	0.755
	0.764	0.759	0.817	0.807	0.862	0.849