Hello,

I am testing for differentially expressed genes via the glmLRT function, and I have some basic questions. I bet I am not the first one to ask, but I couldn't find the answer here.

Are the fitted.values the data normalized to a negative binomial distribution? More specifically, why is fold change the coefficients divided log(2), instead of the traditional logarithm of the ratio between the means of the normalized data? Maybe what I don't understand is what the coefficients are.

Thanks!

Actually you are the first to ask a question quite like this.

The meaning of all the edgeR output is described in the documentation and case studies. If you have two groups, then the logFC output by edgeR will indeed be the log2-fold expression changes between groups. More generally, the meaning of the coefficients obviously depends on your experiment design and on your design matrix.

edgeR is a large package and you will generally learn more from reading the documentation than by trying to guess what it is doing from nosing around the internal code. The reason for dividing by log(2) in the lmFit code is simply to convert from natural logs (used internally) to base 2 logs (used for all the output).

There are no "pseudo-counts" in edgeR so the title of your question is unclear.