Figure 4

Type I errors for all methods at a pre-specified significance threshold. The box plots summarise results across three independently obtained simulated datasets for three different simulation settings. In all cases, exactly zero genes were truly differentially expressed. To make possible a comparison between methods that return P values (DESeq/DESeq2 and edgeR) and those that return posterior probabilities (DGEclust and baySeq), we report the Type I error rate at a relatively high false discovery rate, FDR = 10%. In all cases, DGEclust maintains a minimal Type I error rate, particularly for small sample sizes (n=2). DE, differentially expressed.