Hi,

I've two factors in my experiment design : group (Normal, Low, Middle, High) and localization (in, out). We would like to perform DE analysis to extract DE genes based on localization (in or out).

My first idea was to use ~ group + localization as model and then results(dds,contrast=c("localization","out","in")) to extract the DE genes. However there are previous evidences that the localization may have an impact on the group i.e. a sample is more prone to be in Middle and High if localization is in.

Knowing that, should I use ~ group + localization + group:localization or merge group and localization together and perform ~ group_localization and then run DE analysis on each group e.g. for Middle : results(dds,contrast=c("group_localization","Middle_out","Middle_in")) ?

Thank you

Either approach works to split your localization by group. In the vignette section on interactions we describe how these are equivalent approaches. The latter is often easier for users to work with as you can directly compare groups in a pairwise fashion with contrast.