DESeq2 - Pairwise comparisons in multifactor experiments

[JakobW]

Hi SeqAnswers, I have a question about how to use DESeq2-factors properly for pairwise comparisons in multi-factor experiments.

I have been looking through the threads on the forum for some time now, and learned a lot in the process. But I still have one question I haven't managed to find a clear answer on.

We have RNA-seq data from 36 samples. They come from two different species, three different timepoints, are either treated or untreated and we have three replicates for each sample.

My condition-matrix looks like this (but with three replicates for each datapoint):

Sample Species Treated Timepoint
1 A Treated 1
2 A Treated 2
3 A Treated 3
4 A Untreated 1
5 A Untreated 2
6 A Untreated 3
7 B Treated 1
8 B Treated 2
9 B Treated 3
10 B Untreated 1
11 B Untreated 2
12 B Untreated 3

We are interested in looking at differentially expressed genes in each species for each timepoint: Differentially expressed genes in treated vs untreated A at timepoint 1, timepoint 2 and so on.

One approach to this which is mentioned in some threads is to create another column where the factors are merged:

all_factors
ATreated1
ATreated2
ATreated3
AUntreated1
...

Using the design:

~all_factors

And then perform the desired comparison using a contrast matrix:

res <- results(deseq_obj, contrast=c("all_factors", "ATreated1", "AUntreated1"))

I am wondering if this is reasonable approach, or if I lose information compared to if I had used the design:

~Species+Treated+Timepoint+Species:Treated

(I believe it should include Species:Treated as we expect that one of the species will respond stronger to the treatment, but I might not grasp the interaction fully yet.)

If this second design is preferable I have yet to figure out how to contrast it properly. It seems for me that the standard way of contrasting DESeq2-results only supports comparison of two levels in one factor (for example, all treated samples vs all untreated samples).

Any input on this would be highly appreciated! Both concerning how to approach this and whether I have misunderstood anything about the analysis. Thank you.