[BioC] loop design with limma

[Gordon K Smyth]

> Dear all,
>>From what I understood to when I read-- Kerr, M. K., and G. A.
> Churchill 2001 Statistical design and the analysis of gene expression
> microarray data. Genet Res 77(2):123-8--, to analyze loop design
> correctly one needs to assess the dye effect and the spot effect. In the
> model M+AD+(AD)+G+(AG)
> (DG)+(VG), the interaction of the arrays and genes(AG) and the
> interaction of dyes and genes (DG) should be 0. Thus, analysis based on
> this model should use the measurements of each channel separately rather
> than using the channels ratio.

No, there is no need for this.  limma is designed to analyse "loop" designs and similar, but it
based on more recent work than the paper you cite, which is from 2001.

The early work that you cite proposed to fit a global anova model to all genes simultaneously. 
This approach makes very strong assumptions, e.g., all genes have the same variabilitiy, and
provides no direct means to test which genes are differentially expressed.  Genewise analyses
using log-ratios are more flexible, make fewer assumptions and more directly address the questions
of interest.

> In limma, after normalizing, we left with
> M and A values and if what I described above is correct I don't
> understand how to analyze loop design in limma.
> Is there a way to get the normalized channels rather than the
> normalized ratios?

Have you looked the Limma User's Guide?  That has an entire section on separate channel
normalization although, as explained above, this is not required to analyse loop designs.

Gordon

> Thanks,
> Ron