EdgeR question and looking at significant differences in fold changes.

[GreboGuru]

Hello all,
I want to use a GLM in EdgeR on raw fragment counts generated from cuffdiff. I have 4 groups (A,B,C,D) with 4 biological replicates each. I want to determine if there is a significant difference between the fold change between A/B and the fold change C/D. So I want to look at significant differences of significant differences. I am thinking of using the following command in EdgeR:

> my.contrast <- makeContrasts((A/B)-(C/D), levels=design)
> lrt <- glmLRT(fit, contrast=my.contrast)

Can anyone give me a sanity check on this?

[dpryan]

Correct me if I'm misunderstanding, but can't this be restated as looking for interaction effects in a stereotypical 2x2 factorial design? In effect, that seems to be what you're trying to do.

[GreboGuru]

Thanks for the input!
That was also my first thought, to run a two-way ANOVA where:
y = diet + genotype + diet*genotype
and then any interacting effects would be cases where the diet had a significantly more pronounced effect on one genotype than the other. EdgeR should be able to handle this as well,right? There a better way to do this than an ANOVA?

[dpryan]

Probably a typo, but you mean "y=diet*genotype" or "y=diet+genotype+diet:genotype" (both are the same), otherwise you fit with redundant coefficients :P

edgeR can definitely handle situations like this (as can DESeq and a few others). Have a look at the user guide for edgeR for some examples of this. I've used similar models to look at drug:genotype interactions and diet:gender interactions (mostly with DESeq, but edgeR is really similar). This is also the standard way to look for these effects, so whomever end up reviewing the paper including this data should be comfortable with this approach.

[GreboGuru]

Kind of a typo - I was thinking of how it would be done in SAS. Thanks again for the tips. Alles klar!