Dear,

I have been using edgeR for many years, and there is a problem that always confuses me. Just have a look at this experimental design ( control: replicate1, replicate2, replicate3; treatA: replicate1, replicate2, replicate3; treat B: replicate1, replicate2, replicate3). There is no correlation between treatA and treatB.

I wanna to obtain DEGs between any combination: treatA vs control, treatB vs control, treatA vs treatB, here are two methods:

1) construct a DEGlist consisting of control, treat A, treat B and get the DEGs with coef=2, coef=3 and contrast=c(0,-1,1) 2) construct three DEGlists consisting of (control, treat A), (control,treat B) and (treatA, treat B) respectively and get the DEGs respectively.

In short, my problem is that can I put all the irrelevant samples into one DEGlist and perform differential expression anlysis. Will be the results between the two methods the same ? I think that they would be different, which one is more reasonable ?

If these samples were obtained from the same experiment, it would be better to include all of them (your method 1) in your analysis. This will give you more degrees of freedom in dispersion estimation and more power for your statistical tests.