[BioC] edgeR - coefficients in 3-factor experiment, complex contrasts and decideTestsDGE

[Gordon K Smyth]

On Tue, 18 Oct 2011, Gordon K Smyth wrote:

>> From: Emanuel Heitlinger <emanuelheitlinger at googlemail.com>
>> To: bioconductor at r-project.org
>> Subject: [BioC] edgeR - coefficients in 3-factor experiment,	complex
>> 	contrasts and decideTestsDGE

>> In glmLRT giving simple coefficients would compare the complete model 
>> to a model removing one coefficient at a time. From application of glms 
>> in ecology I remember that an interaction effect should not be left in 
>> the model if the main effect is removed. Does this apply here? Should I 
>> compare the full model against e.g. the model minus pop, sex:pop, 
>> eel:pop and sex:eel:pop, when I want to remove condition "pop" from the 
>> model?
>
> Yes, that's true.  To test whether pop makes any contribution to expression 
> changes, you need:
>
>  lrt <- glmFit(d,design,coef=grep("pop",colnames(design)))
>
> etc.  That does a test on 4 degrees of freedom.

Sorry, that should be

   lrt <- glmLRT(d,glmfit,coef=grep("pop",colnames(design)))

Best wishes
Gordon

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}