Fig. 7

Comparison of methods using real data sets. a-e Analysis results based on the ABRF data set. f Analysis with the modencodefly data set. (a) Venn diagram of the number of DE genes identified by the three methods. Numbers in brackets indicate the eFDR. The specific gene sets of aFold alone or DESeq2 and Voom combined are indicated by roman numbers I and II, respectively. b-c eFDR as a function of different cut-offs of either adjusted p-value (b) or fold change (c) for gene sets I (aFold) and II (DESeq2 and Voom). The two inlets show the results based on all DE genes (rather than the subset of genes). d eFDR (left Y axis) and percentage of detected DE genes (right Y axis) for different quartiles of the data (X axis). Solid lines indicate eFDR under adjusted p-values of 0.05, dashed lines under adjusted p-values of 0.05 and a log fold change of at least 0.5. Red, turquois, and magenta are as in b and c. Grey line and points show eFDR for all genes (including both DEs and non-DEs). Genes were grouped according to expression (q1, q2, q3 and q4 in boxplot). Lines in blue and green show percentages of detected true DE genes across quartiles for gene set I and II, respectively. e-f Type I error rates for the ABRF (e) and modencodefly (f) data sets. Type I error rates are calculated via the number of DEs under p-value < 0.05 divided by the total number of genes