[BioC] Limma, model with several factors

[Ryan C. Thompson]

Ingrid,

It is possible to block for any number of factors in limma. However, if 
you try to block for subject in the experiment that you showed, you 
will find that you don't have enough replicates to do so. The 
combination of treatment and timepoint contributes 4 degrees of freedom 
to your model, and subject would contribute 4 more, for a total of 8. 
This is the same as the number of samples in your experiment, which 
means that you have no remaining degrees of freedom. It would be 
analogous to trying to calculate the standard deviation of a single 
sample. Your best option is probably to block only for center and not 
subject.

-Ryan

On Mon May 20 01:01:08 2013, Ingrid Dahlman [guest] wrote:
>
> I have carefully read the Limma users guide but still have not sorted out how I design the contrasts for the following Project.
>
> In want to compare the effect (After vs before) of two different treatments, F and P. The study was carried our in two different centers. This can be illustrated as follows:
>   Subject center treatment timepoint
>   1            1         F          B
>   1            1         F          A
>   2            1         P          B
>   2            1         P          A
>   3            2         F          B
>   3            2         F          A
>   4            2         P          B
>   4            2         P          A
>   I want to compare (F.A-F.B)-(P.A-P.B), blocking for subject. However, in addition, I would like to block for center. I.e. the center is like a batch effect.
> Is it possible to block for two factors, suject and center, in the same test in Limma?
>   /Ingrid
>
>   -- output of sessionInfo():
>
> See section 8.7 in the user guide.
>
> --
> Sent via the guest posting facility at bioconductor.org.
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor