Hello,

I am attempting to use edgeR (3.12.1)'s glmTreat() function to identify genes that are significantly up-regulated by >2-fold in one group over another. My understanding from Sections 2.12 and 4.4.8 of the edgeR User's Guide is that the log2 fold changes for each gene will be tested against the absolute value of glmTreat()'s lfc parameter, which would output differentially expressed genes that are both up- and down-regulated. Is it possible to perform a more specific test in edgeR to identify just the significantly up-regulated genes above a lfc threshold, instead of extracting the subset of up-regulated genes post hoc?

Thanks in advance for your help.

Is this really necessary? If you're not interested in the downregulated genes, just ignore them. Doing anything fancier would be more trouble than it's worth. If you really want to do extra work, here's some possibilities:

• For all upregulated genes, divide the p-value by 2, while for all downregulated genes, replace the p-value with 1 - p-value/2. This effectively converts a p-value of a two-sided test to that of a one-sided test. Obviously, this improves power for the upregulated genes, as you lower their p-values.
• Subset the genes to contain only those that are upregulated, and apply the BH method on the subset. This improves power by only correcting across the subset of genes that are DE in the direction of interest.

Both strategies are based on the idea that the p-value is independent of the sign of the log-fold change in a LRT. However, I can't recall what the proof was; something about the root likelihood ratio being normally distributed...

Edit: It's probably better to go with the second option above. The true one-sided p-value from TREAT will be close to the two-sided p-value due to the asymmetry in the tails when a gene is up-regulated, so dividing by 2 will lead to lower-than-correct values.