I am using DESeq2, but unsure whether or not I am using the right design. I could not find here anything similar.
subject: A, B, C, D, E, F cohort: ctrl, treatment time: 0, 1
As you can see subjects belong to either cohort (ctrl or treatment) and each subject is sampled at multiple time points.
Is there a way in DESeq2 to "block" for subjects? In my case (my data) base counts can vary substantially among subjects, and I am trying to find genes that behave differently in time, depending on the treatment (cohort).
Q1: Would this design work for me?
design(dds) <- ~ time + cohort + time:cohort dds <- DESeq(dds, test="LRT", reduced = ~ time)
Q2: How can I implement the "blocking" of subjects? If you are going to suggest duplicateCorr from voom/limma, at what point of the DESeq2 analysis should I do that?
Q3: Do I really need to exclude samples when cohorts are unbalanced (e.g. 30 subjects vs 24 ?) It would be better if DESeq2 could do this, for instance by randomly excluding 6 subjects and re-running this analysis multiple times, rather than me manually discarding data from 6 samples... not sure if possible..?
extra Q : Would I also be able to find genes going up/down the same direction in both cohorts, when the slope is higher in one cohort than in the other? would those also be detected as DE?
Thank you for taking the time! Dany