Box plot of the CV of p-value histogram of permuted microarray datasets. PermLab: Breast cancer dataset with class labels randomly permuted. By randomly permuting class labels, all genes become non-differentially expressed but the gene-gene correlation intact. PermAll: Randomly permuting all expression values between the 111 samples and 9993 genes. By randomly permuting expression values all genes become non-differentially expressed and independent with each other. The 2 types of permutations were performed 100 times each, and the p-values were split into 20 bins between [0, 1] with the width of each bin being 0.05. The horizontal line represents the expected CV when genes are independent.