Hi,

I am using edgeR for RNA-seq analysis and wanted to check: Is it acceptable to include Gender (Male and Female) as a factor in the design when using the filterByExpr function, in addition to the two-group comparison (Group A vs Group B)?

I include Gender as a covariate in the design (during dispersion estimation and fitting the model) like design.2 (see below). I am not sure during the filtering stage should I use design.1 or design.2 ?

Currently, I use the following approach to filter lowly expressed genes:

design.1 <- model.matrix(~0+Group)

keep <- filterByExpr(y, design.1)

table(keep)


design.2 <- model.matrix(~Gender+Group)

keep <- filterByExpr(y, design.2)

table(keep)

Either of these choices is perfectly acceptable from an edgeR analysis point of view, it is more to do with which is more relevant for your scientific purposes, and which contrast comparison you are intending to make downstream.

I am guessing that Gender is more a nuisance factor in your experiment, and your primary interest is to compare groups. In that case, I would just use filterByExpr(y, group=Group). No need to make a design matrix at all.

In edgeR v4, the requirement for filtering is greatly reduced. In edgeR v4, you could take a different and more inclusive approach. You could specify the minimum number of samples that you would like a gene to be expressed, in order for a DE result to be of scientific interest to you. That might be the minimum group sample size, or it might be even less. Then just keep genes with a positive count in that many samples. I would recommend this sort of approach particularly for large-scale studies with lots of samples.