[BioC] Batch effects and linear models

[Malard, Joel M]

This is a naïve question. Is there some BioConductor package to estimate mixed models on Affymetrix data?

I have a 2 x 2 design, treatment versus control at two labs. There are strong (lab) batch effect that can be seen by clustering the data. So far I have computed RMA (and GCRMA) values for each lab separately, normalized across arrays using limma's normalizeBetweenArrays (quantile) and generated lists of potentially differentially expressed genes via lmFit, eBayes and TopTable. I would like to estimate how much of a batch effect remains. 

>From my numeralogist roots, I know that a mixed model across all arrays need not give the same estimates as a linear model of the residuals from linear models derived for each lab. Any pointer on how other people handle batch effects would be most useful and most welcome.

With best regards to all,

Joel M. Malard, Ph.D.
Scientist IV
Pacific Northwest National Laboratory
Tel: +509 372-6539	Fax: +509 375-2604
Battelle Boulevard, PO Box 999
Mail Stop K6-08
Richland, WA 99352

"I love the audacity of those who have everything to lose from it; the moderation of those who have nothing to gain from it." Rostand, Jean (1894-1977)



	[[alternative HTML version deleted]]