I know this question has been asked multiple times but my set-up is a little weird. I am comparing gene hits from a transposon screen and have 7 different groups of 3 replicates each and would like to pairwise comparisons between every group. I tried combining all the data together into a single
dds object, but the unadjusted p-value histogram from this was not correct (see Scenario D in http://varianceexplained.org/statistics/interpreting-pvalue-histogram/). The data is very zero-heavy and I tried filtering out rows many zeroes but this did not completely solve the issue. Additionally, I checked the variance on a PCA plot and saw that 4 of my 7 groups had high intergroup variance. I switched to making running DESeq multiple times with two groups for each run and got much cleaner p-value histograms (Scenario A in the link above). Based on what I've read, I think this is a legal strategy but I want to make sure. Please let me know if this needs more clarification or if you folks have any advice to add.
Thanks and thanks to Michael Love for the great tool!