The ‘un-shrunk’ partial correlation in Gaussian graphical models

BMC Bioinformatics

Table 1 Area Under the Receiver Operator Curve (AUROC)

p = 100	‘un-shrunk’	‘shrunk’	gLASSO**
n = 10	0.984772*	0.957152	0.995724
n = 20	0.992656*	0.960648	0.995772
n = 30	0.994636*	0.961956	0.995928
n = 40	0.995132*	0.963972	0.99576
n = 50	0.99556*	0.965676	0.995932
n = 60	0.995804*	0.968264	0.995788
n = 70	0.995896*	0.970764	0.995896
n = 80	0.995956*	0.973992	0.995964
n = 90	0.996052*	0.975976	0.99588
n = 100	0.996044*	0.977428	0.995896

The table provides the average AUROC score across 25 data sets for a network with p = 100 nodes with 1% of true positives edges and varying sample size n
*Two-sides t-test between the ‘un-shrunk’ and ‘shrunk’ methods are statistically significant at 0.05
**STAR model selection is used from the R package ‘huge’ [15, 19]

ISSN: 1471-2105