To all DESeq2 experts or biostatisticians out there:
Say I have an RNA-seq experiment with three conditions: A, B, and C. I want to know which genes are differentially expressed specifically in A; that is, different between A vs. B, A vs. C, but not B vs. C.
Solution 1) The naive approach of course is to perform pairwise comparisons and pull out genes specific to A via set overlaps of the three resulting gene lists. What I don't like about this is that I have to impose two arbitrary cut-offs: one to establish differential expression (say FDR < 0.01, |log2FoldChange| >= 1) and one to establish no differential expression (say FDR > 0.5, |log2FoldChange| < 1). Another disadvantage is that I am stuck with three different FDR and log2FoldChanges that I am not sure how to combine into a final gene ranking, for example for pre-ranked GSEA.
Solution 2) An alternative is to compare A with all non-A, but this would not guarantee me non-differential expression between B and C. Furthermore, I guess I sacrifice power by withholding the information that 'non-A' is actually composed of two conditions (B+C).
After implementing both solutions they still feel like a hack. Can this problem be solved in one nice integrated statistical framework that also allows me to nicely rank my resulting DEGs by statistical significance?
I read about contrasts and likelihood ratio tests in the DESeq2 vignette, but could not come up with an answer so far.