ROC curves of simulated RNA-seq data in different scenarios. ROC curves in the scenarios of baseline background noise (a); Increased background noise (b); Increased background noise but only with highly expressed genes above 75% quantile of intensity (c); Increased background noise but only with lowly expressed genes below 25% quantile of intensity (d); Comparison with other differential expression analysis methods (under increased background noise) with either 3 samples per condition (e), or 6 samples per condition (f). The diagonal black dash-line indicates performance under random events (0.5 AUC). The light blue line in (b,c,d) indicates the theoretically optimal ROC when the background noise can be estimated exactly. Simulation for (a, b, c and d) was carried out 100 times. Simulation for (e and f) was carried out 10 times. 3 replicates per test condition, with 10% DEGs and 1.5 fold change were used for (a, b, c, and d).