False positive rates and ROC curves of CSA for different parameter settings. (a) False positive rate of CSA with different parameter settings at different FDR levels. (b) ROC curves of mean function and ratio function of low/high and high/low simulated data. (a) The plot clearly shows that the estimated FDR can well control false positive rate of CSA. Both scoring functions with graph permutation reach low false positive rate when applying a reasonable FDR cutoff (FDR < 0.05). (b) The ROC curves suggests that the ratio scoring function reaches better true positive rate at the expense of a similar gain in false positive rate on datasets that contain few highly correlated regulatees.