DESeq2 - paired formula design with replicates

[bryand]

Hi all,

I know that this is a frequently asked question, but I still haven't quite grasped how to build the formula design for a multifactorial paired study. My sample table is as follows:

Has.disease PatientName Replicate Treatment Individual
N Control 1 Ctrl 1
N Control 1 Epi 1
N Control 2 Ctrl 2
N Control 2 Epi 2
N Control 3 Ctrl 3
N Control 3 Epi 3
Y Patient1 1 Ctrl 1
Y Patient1 1 Epi 1
Y Patient1 2 Ctrl 2
Y Patient1 2 Epi 2
Y Patient2 1 Ctrl 1
Y Patient2 1 Epi 1
Y Patient2 2 Ctrl 2
Y Patient2 2 Epi 2
Y Patient2 3 Ctrl 3
Y Patient2 3 Epi 3

I'm interested in identifying those genes which have a different expression profile between disease/not-disease patients after treatment with Epi. I finally settled on a design formula of: ~Has.disease + Has.disease:Replicate + Has.disease:Treatment (per the vignette), but I really don't know 1) What exactly the formula is measuring, and 2) How I should set my corresponding contrasts.

As an aside, I've had a very hard time understanding when I should be using the : or the * interaction term in the design formula - I've had no luck with the DESeq2 documentation or on the online forums with finding a good guideline.

[jp.]

ops.. I also want that DESeq2 treats samples as replicates but it doesnt as attached

[Michael Love]

This means you want to test the interaction of treatment and disease, controlling for the individuals within each disease group.

In R formula this is:

~ disease + disease:replicate + treatment + disease:treatment

Make sure that replicate is a factor, not a numeric. Make sure you set the reference level of disease and treatment using relevel() or the levels argument to factor() (see vignette).

And you test the final term, disease:treatment. You can do this with:

dds <- DESeq(dds, betaPrior=FALSE)
res <- results(dds)

The betaPrior=FALSE is recommended for designs with interaction terms. This will happen by default in the next release (Oct 2015).

The interpretation of the interaction term is the additional effect of treatment in the disease group compared to the no disease group.

If you are not familiar with interaction terms, and want to learn more, it's best to have a local statistician explain them (this is not specific to DESeq2 but a general concept in statistical analysis).

[jp.]

Thank you Michael for solving my trouble
you are the leader