Dear all,

I am new to RNA seq data analysis. I have been trying to do some DE analysis using DESeq2. I am comparing two conditions, and I expect two genes X and Y to differ in these conditions. When I put condition 1 as reference, the MA plot looks as figure A, and when I do lfcShrink using apeglm, it looks like figure B, my genes of interests still there. When I put condition 2 as reference, the MA plot is the figure C. However, this time after shrinkage (figure D), the lfc of one of my genes of interest goes to almost zero. Can you please clarify why this is happening? Best![enter image description here][2]

Figure

I think that the adaptive prior of apeglm is not working well for this dataset. It looks like you have many genes consistent with LFC=0 and then two very large effects. Obviously, this is the kind of situation we were trying to accommodate when developing apeglm, but we use a Cauchy prior, and it looks like this needs a prior with even more extreme "kurtosis" (more of a spike and slab shape). Can you try type="ashr", which fits a flexible mixture prior?