Hi everynone!

I am using DeSeq2 to generate a list of differentially expressed genes when different treatments are applied (A vs. con, B vs. con and C vs. con). I am interested what kind of p-value correction method should I use. I know that padj calculated by DeSeq adjusts the value based on the number of tested genes, but should the correction method consider that I also have 3 different comparisons (in fact, for each gene I have 3 tests, one for each comparison)? I would like to emphasize that treatments weren't compared at any point, just treatment vs. control.

Also, I would appreciate it if you could provide examples of scientific papers where this issue was addressed and the correct correction method was applied to a real scientific problem and data.

Thank you all very much!

I don't typically correct across a few pairwise comparisons, but instead just consider that each results table controls its own FDR.

You can see how often, under the null, would BH give you any false rejections across multiple results tables, it's not a huge concern:

> replicate(10, sum(p.adjust(runif(1000)) < .1))
 [1] 0 0 0 0 0 0 1 0 0 0