Hi

i am analysisng a dataset w two classes 14 samples, balanced

when i run classic or glm and select genes by p value i get more or less the same result

when i select genes based on fdr i always get 0 genes from GLM

is that ok or what that means?

mario pescatori

The short answer is that any multiple testing correction will adjust the p-values to larger values. So, the number of genes with p-values below 0.05 (or any other threshold) will generally be smaller than the number of genes with adjusted p-values (i.e., FDR, via the Benjamini-Hochberg method) below the same threshold. In your case, the adjustment of the p-values from the GLM framework in edgeR means that you get no DE genes at your FDR threshold. Oh well, that's life; you shouldn't be selecting genes on the raw p-values anyway.

Now, for the long answer; what your post really seems to be asking for is the differences between classic and GLM edgeR. Presumably you detect some DE genes at a particular FDR threshold with classic edgeR, and no DE genes with GLM edgeR. I'll be blunt; there are no circumstances in which I would use classic edgeR over the GLM mode. GLM edgeR (and in particular, the quasi-likelihood framework) contains a number of improvements in terms of speed, accuracy and flexibility. The real issue is whether you set up your GLM analysis properly, and that's not a question that can be answered without you showing your experimental design and some code.

P.S. Please use the "add comment" button to reply to this post, not the "add answer" button.