nhaus - This question has been asked many times both here as well as on other fora, for instance the bioconductor forum.
Mike Love (the author of the DESeq2 software) has probably answered this question 50-60 times, if I had to guess, so you can find a lot on this subject by googling.
However, to guide you a bit more, I would encourage you to consider the following resources:
- The DESeq2 Bioconductor page, which contains many of the rest of the links below.
- The DESeq2 Vignette; in particular, the subsection on interactions.
- Both the DESeq2 Reference Manual and the Limma Reference Manual.
In addition to these, I want to mention a few terms that you can search and read about that you may not come across in searches like the above...
Background Reading (the basics):
- Go to Google or Wikipedia and search first for "Multiple Linear Regression" then for "Main effects versus interaction effects".
If you want to get more technical:
- To really (really) understand what is happening in a given model or with a given software, you need to understand what is meant by the terms
general linear modeling and then
generalized linear modeling.
- Once you have familiarity with those two, note that DESeq2 is a form of GLM called
- Once you have familiarity with 1 & 2, re-visit the term "
Sums of Squares". Not all softwares calculate Sums of Squares (SS) in the same way (SAS and R default settings differed in this regard the last time I checked). There are different ways in which to calculate SS. To provide a quantitative/precise answer to the question you've asked above, you'd need to know a bit about how that's being done.
I know this may seem like minutiae or a pedantic point. Maybe it is, but it can definitely influence your results. As an example of why this could be important to you, consider:
Gene Expression ~ Genotype + Treatment Group + Age
Gene Expression ~ Treatment Group + Age + Genotype
Depending on the type of SS being used (Type I, Type II, Type III etc.) simply the order of the terms alone may or may not influence the effect size estimates (beta, odds ratio) generated ...
OK, final comment. Above, you asked, which is "correct"? Consider the example just above. Which is correct? Well, it depends on exactly what you are trying to study, and exactly what question you are trying to ask.
If you have a chance to review the above and are still unsure which is best for you, please let me know.