I have been analysing an RNA-seq data set with multiple conditions (injury and treatments), and have generated many DE gene sets using the DESeq2 lfcShrink() function with the apeglm method.
As the PI is preparing to publish the results he has asked that one of the volcano plots should be mirrored, so I recalculated the fold changes with the conditions reversed and replotted. I was surprised to find that the fold changes are not the same with the sign reversed, as I had expected they would be. I repeated in a vanilla DESeq2 session and confirmed that whilst the inverse comparison using the results() function results in a sign switch for the fold changes, using lfcShrink modifies the fold change estimates depending on which condition is used as the baseline.
Can anyone explain why this happens when using lfcShrink/apeglm? I had naively expected the changes to be mirrored as happens with the non-shrunken fold changes. (I'll also go back and re-read the apeglm paper to see if I can spot the reason in there).
Thanks,
Chris
