g

p

le the t test p value was 9.0e−8.

(a) (b)

he expression profile and the p values of seven tests. (a) The up-regulated DEG

wn-regulated DEG.

cover heterogenous DEGs for a cancer data set

n tests were used to discover heterogeneous DEGs for the breast

agnosis data (GDS3139) [Tripathi, et al., 2008]. For all tests, the

p value for this data set was 0.01. Figure 6.22 shows the

ons of the p values of the COPA, OS, ORT, MOST, LSOSS and

dels constructed for this data set. The distributions of the p values

OS, ORT, MOST and LSOSS models were questionable. As

earlier, the density of the p values should be a mixture of a

distribution for the non-DEGs and a Gamma distribution for the

usher, et al., 2001; Storey and Tibshirani, 2003]. Based on this

nly the p value distribution of the DOG model was reasonable.

e 6.23(a) shows the correlation analysis plot of the negative base

ithm of the p values of the models using the seven tests for the

t can be seen that the DOG model was highly correlated with the

OS model was highly correlated with the ORT model and the

odel was highly correlated with the ORT model. Figure 6.23(b)

Venn diagram analysis of the detected DEGs for the breast cancer

by the DOG, COPA, OS and ORT models. It shows a similar

ound in the correlation plot that the COPA model was more