I'm wondering if there's a way to leverage the log2FoldChange and lfcSE (log2FoldChange standard error) in order to calculate whether fold changes between genes are significantly different. (i.e., in the graph below, is there a way to determine whether this difference in fold change between these 2 genes in this experiment is significantly different?)
It seems like all of the information is there to perform some sort of statistical test, though I'm wary of back-calculating the variance from the lfcSE in order to use a formula for, say, a t-test.
Thanks!
Hi Micheal, do you mind expanding on how you would put the counts from both genes together so we can get the differences between the fold changes of two different genes?
This also raises a question of how do we compare levels of different genes against each other in a group of samples? This is obviously limited by the problem of gene length and GC bias, but after correction for that (EDASeq), could we then use the DESeq2 model to make this comparison?
While it seems possible (having a gene term, and then testing the interaction between gene and condition) I don't see the point, so I'm probably not going to work out a software implementation for this. I think just showing the two coefficients and their SE or posterior intervals is preferable to a null hypothesis test.