Fig. 3

DGE simulation study results for the limma-voom analysis with five replicates in every treatment × time combination. a FDR and OFDR control for the conventional approach (blue), the stage-wise method proposed in this manuscript (green) and the stage-wise method from Jiang and Doerge [13] (orange). The false discovery proportion (FDP) is assessed in 30 simulations, which allows us to evaluate the FDR as the mean over all FDPs. The conventional method controls the FDR over all hypotheses but is too liberal on the OFDR, and the Jiang and Doerge method seems to be overly conservative in all scenarios. The stage-wise procedure we propose controls the FDR over all hypotheses; however, this is generally not guaranteed. As expected, it controls the OFDR on all significance levels. Compared to the conventional approach, the fraction of null genes (genes with no effect whatsoever) among the OFDR false positive list is lower for the stage-wise testing procedure proposed in this manuscript, which shows that it is advantageous in terms of efficient biological validation of the results. b False discovery proportion-true positive rate (FDP-TPR) performance curves for the treatment × time interaction effect based on the first simulation. The three points on the curves represent nominal FDR cut-offs at 1%, 5% and 10% and are filled if the empirical level is below its nominal level. The proposed stage-wise method boosts power for the interaction effect through the enrichment of interaction genes in the screening stage. The Jiang and Doerge method enriches for fewer genes as compared to the Heller method, because it has to split the FDR between its two stages. Furthermore, the Jiang and Doerge method is very conservative since it only allows control on the upper bound of the FDR across the hypotheses