Differential expression analysis with DEseq2 : multiple groups comparison

Question

[guillaume.rbt]

Hi all,

I think this question may have been already asked before, but I have not been able to find an answer, or I did not fully understand the answers, as I'm a beginner in DE analysis.

Considering the following RNAseq dataseq, imported in DEseq2 : 12 samples, with 2 conditions, and 3 replicates fore each combination of conditions

NT: normal temperature

ST : stress temperature

NW : normal watering

SW : stress watering

> colData(dds)
                      temperature    watering
                     <factor>        <factor>
sample1.1     NT                    NW 
sample1.1     NT                    NW 
sample1.1     NT                    NW 
sample2.1     NT                    SW
sample2.2     NT                    SW
sample2.3     NT                    SW
sample3.1     ST                    NW
sample3.2     ST                    NW
sample3.3     ST                    NW
sample4.1     ST                    SW
sample4.2     ST                    SW
sample4.3     ST                    SW

I would like to do a "all vs all" comparison between all the groups of replicates,but i'm not really sure what design should I use for that.

By this I mean that I would like to have a list of all genes that are differentially expressed between all combination of conditions NT/NW, NT/SW, ST/NW, ST/SW. (sample1 vs sample2, sample 1 vs sample3, sample1 vs sample 4, sample 2 vs sample 3 ... etc... )

Thanks for any of your input.

Answer 1

[Devon Ryan]

You say you want all of the pairwise comparisons, but I question whether that makes the most biological sense, after all, you designed your study to look at the effects of temperature and watering (and the interaction of the two), so that seems like the biologically most tractable way of doing the analysis.

Anyway, if you really do want the pairwise comparisons, then add a new column to your column data:

colData(dds)$group = paste(colData(dds)$temperature, colData(dds)$watering, sep=".")

and then use ~group as the design. You can then specify each of the comparisons with contrast= in lfcShrink().

[guillaume.rbt]

Thanks for the info anf for the advice. I'm curious, what would be a more relevant design to study the interaction between temperature and watering in this case according to you?

[Devon Ryan]

~temperature*watering, which would give you the interaction effect directly as a coefficient. I find factorial designs like this to be generally more biologically interpretable than the group design used in pairwise comparisons, since then you'll notice differences due to changes in watering in the different temperatures and won't be able to easily get at what generally changes with watering (a more interpretable question IMO).