I'm working on a RNA-seq analysis project and I have come across a statistics problem. I have 3 groups and I need to look for differential expression for each pairwise comparison.

Do I control the FDR separately after each pair, or do I control the FDR for the entire analysis at the same time?


2 Answers 2


I just asked the same question on Bioconductor.


You could do an F-like-test for each gene, then you would only be doing 1000 tests. In edgeR, you simply give both pairwise comparisons to glmLRT() or glmQLFTest() simultaneously and it will do a combined test.

Alternatively, if you want to keep the two comparisons separate, see ?decideTests for the various ways in which you can further adjust the p-values across comparisons. Usually people just adjust each comparison separately.

So you may just adjust each comparison separately if your software is not edgR and doesn't combine your tests.


If you control the FDR for the entire analysis, the number of FD can be higher than you wish in a pair and lower in another, so that it is control at your chosen level for the entire. Consequently, you have to control the FDR separately to ensure it is well controlled for each pair.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.