[BioC] LIMMA: donor effect

[Gordon K Smyth]

Dear Som,

The usual way to deal with the donor effect is to include a factor for 
donor in the design matrix.  This ensures that all your comparisons are 
made within donor:

   design <- model.matrix(~Donor+Dose*Time)

Usually I would prefer to correct for the batch effect in the design 
matrix, rather than pre-correcting with ComBat, because it ensures that 
the degrees of freedom reflect that fact that batch correction the has 
been done.

Best wishes
Gordon

> Date: Fri, 22 Apr 2011 13:49:59 -0400
> From: somnath bandyopadhyay <genome1976 at hotmail.com>
> To: <genome1976 at hotmail.com>
> Cc: bioconductor-bounces at r-project.org, bioconductor at stat.math.ethz.ch
> Subject: [BioC] LIMMA: donor effect
>
>
> I am trying to analyze microarray data for a project where whole blood 
> was collected from 4 different donors and was treated with a compound at 
> 2 time-points at 4 different doses. I am interested in genes that are 
> differentially regulated with dose, with time and also dose*time. When I 
> run a PCA on the entire data, the separation of groups is mainly by 
> donors. How do I run a limma analysis (factors, design matrix) to get to 
> the genes of interest (mentioned earlier)? I am not interested in genes 
> that are differentially expressed across donors. Any suggestion would be 
> greatly appreciated.
>
> Thanks,
> Som.
>
> Would it be O.K. to use ComBat to get rid off donor effects by treating 
> each donor as a batch?

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