I've been reading up on corrections after DESeq2 and talking to other bioinformaticians and they say that you should adjust for all the comparisons that you made after running DESeq2, or adjust within each comparison. And DESeq2 automatically includes the adjusted p-values that account for multiple comparisons.
My question is this, how do we adjust the comparisons for either case?
You have already noted that DESeq2 automatically adjusts for the multiplicity within each comparison. So I presume you are not asking how to do that. There is an argument that could be made that you should adjust all of the comparisons made in an experiment (so for example, if you had three contrasts, then you would adjust for 3*(Number of genes), rather than simply adjusting for the number of genes separately, three times).
There are even two (IIRC) methods in the limma package intended to make that second multiplicity adjustment. One way is to just take all of your p-values, put them in a vector, and then use p.adjust to adjust them. And then put them back where they belong. But you know what? I would be shocked (shocked, I say!) if more than like 5% of analysts do that on the regular, let alone even once. That's because an RNA-Seq experiment is not like a clinical trial or an Epidemiological study where the results are meant to be inferential and stand on their own. Instead, an RNA-Seq experiment is usually intended to find a plausible set of genes that may be affected in the experimental conditions you are studying. And ideally those genes or pathways identified would be explored further at the bench. If we were doing inference, we would be like the GWAS people and using a Bonferroni. But we are not.
So long story short, just use the adjusted p-values you get, and call it a day.
