[BioC] DESeq /edgeR design problem for a small RNASeq

Simon Anders anders at embl.de
Tue Jun 4 11:52:58 CEST 2013

Hi Eduardo

Thanks for the exaplanation. I now see what yiu are heading at.

On 04/06/13 11:01, Eduardo Andrés León wrote:
>    (MTsD - MTcD) - (WTsD -WTcD) = Real effect of the overexpression.

Yes, this is precisely what is called an "interaction effect" and 
described in R's formula notation with ":" operator.

So, you use, as full model:

full_model <- fitNbinomGLMs( cds, count ~ mutation + drug + mutation:drug)

and as reduced model

reduced_model <- fitNbinomGLMs( cds, count ~ mutation + drug )

and then compare the two:

pvals <- nbinomGLMTest( full_model, reduced_model )

(As an alternative for the notation "~ A + B + A:B", you will often see 
the equivalent abbreviated notation "~ A * B".)

If you deal with such questions more often, it may be helpful to read a 
textbook on linear models and Anova. This is all quite standard stuff.


More information about the Bioconductor mailing list