Evaluation of the appropriateness of the normalization methods on the H data set. For each sample and for each set of replicates, we represent the fold-change distributions (y-axis) for each individual size class (x-axis). Based on the assumption that no significant differences are expected between replicates, a suitable normalization is one that brings all distributions on the 0 line (in log2 scale, this corresponds to equal values in both replicates). For the H data set, the TMM, DeSeq2, and adapted quantile normalization fulfill this criterion for all samples.