Peersim [RNA-Seq]

In stress experiments, for instance, the presence of differentially expressed genes exclusive to combined stresses is frequently interpreted as evidence of non-additive effects: the effect of combined stresses is different from the sum of the effect of single stresses.

Using a multi-stress dataset with 22 replicates, we here argue that this reasoning is likely to be false because of the inherently low statistical power of most studies, which use only a few (often three) biological replicates. While it is well known that low statistical power reduces sensitivity, its detrimental impact on specificity has been largely overlooked. We emphasize the need for careful interpretation of any claim of specificity, regardless of the number of replicates, and plead for improved experimental design and statistical modelling to avoid misleading biological interpretations.