Hi everyone,

I'm new to RNAseq analysis and I have a general question about how to treat genes in DESeq2. Initially I estimated transcript counts in salmon and then imported (tximport) them into DESeq2. Transcripts were converted to genes using the EnsDb.Hsapiens.v86 database, gene counts were estimated and normalized and those became the units (rows) of my DESeq2 analyses. But many of those genes belong to subfamilies of the same gene family or motifs of the same gene, etc. Hence my question is, when would be better, if ever, to group genes by family or motif or any other higher genomic hierarchy? I can see how grouping genes and increasing counts per gene may be statistically beneficial (e.g., less variance, less low-count genes), but is it biologically correct? Any thoughts, guidance, links to previous comments, etc would be highly appreciated.

Then assuming you want to combine genes from the same family or motif and analyze them in DESeq2, how do you do that?

Best regards

Marcos

It's not clear what you mean by 'grouping genes'. If by that you mean something similar to what you did when summarizing the transcripts to the gene level, that doesn't seem reasonable at all. When summarizing the transcripts you do obscure some of the possibly different functions or other biological effects that a different transcript may have, but that is entirely different from assuming that all the genes in a gene family are somehow interchangeable.

A different way of 'grouping genes' is to do some sort of a gene set test, of which there are many. The main goal being to provide evidence that a set of genes that are part of a functional group (gene pathway, etc) are being perturbed. I don't know if there are gene set tests available in DESeq2, although my recollection is that there aren't. You could use things like the goseq package for GO testing, or any of the gene set tests in edgeR/limma (camera, romer, roast, etc), depending on how you want to combine the genes and what null hypothesis you might want to test against.