Fig. 3

mutscan p-value distributions for null comparisons. For each data set, repeated null data sets were generated by artificially splitting the replicates into two approximately equally sized groups. For each such artificial null data set, mutscan (with the method set to edgeR and limma, respectively) was used to fit a model and test whether the log-fold change between input and output samples differed significantly between the two artificial groups. The colored densities represent the individual data splits, while the dark gray density represents the union of p-values from all data splits. Since the groupings are artificial, uniform p-value distributions are expected. While technical differences among the samples, and the low sample size in general, imply that not all comparisons provide exactly uniform p-value distributions, we do not observe a systematic bias in the p-values from mutscan. Only variants with more than 50 counts in all input samples were considered for this analysis. The number of retained features is indicated in each panel