[BioC] Limma and Dye-Swap/Single Channel

[Gordon K Smyth]

> Hello,
>
> I hybridized for two samples both Cy3-Label and Cy5-Label to one chip
> (i.e.
> wtCy3+wtCy5 and mutantCy3+mutantCy5.

Why did you do this?  Your only option now is to do a single-channel
analysis, comparing treatments across arrays, and this is an order of
magnitude less efficient than a direct comparison using Cy3 and Cy5
channels.  Anyway, in limma you could use

MA <- normalizeBetweenArrays(yourRGList, method="quantile")
targetsC <- array2channel(yourtargets)
design <- model.matrix(~factor(targetsC$Target))
cor.fit <- intraspotCorrelation(MA,design)
fit <- lmscFit(MA,design,correlation=cor.fit$consensus)
fit <- eBayes(fit)
topTable(fit,coef=2)

You will have to make sure that yourRGList contains no negative or missing
background-adjusted intensities before using the above.  You could use for
example

RG <- backgroundCorrect(RG, method="normexp")

or similar to ensure this.

Gordon

> I would like to normalize that data, but how?
> I tried to use the array2channel funtion, but how do I use the result?
>
> My targets looking like this:
>    FileName Cy3 Cy5
> 1 e85b.txt  WT  WT
> 2 e87b.txt  WT  WT
> 3 e88b.txt 139 139
> 4 e89b.txt 139 139
>
> After array2channel:
>      Channel FileName Target
> 1.1       1 e85b.txt     WT
> 1.2       2 e85b.txt     WT
> 2.1       1 e87b.txt     WT
> 2.2       2 e87b.txt     WT
> 3.1       1 e88b.txt    139
> 3.2       2 e88b.txt    139
> 4.1       1 e89b.txt    139
> 4.2       2 e89b.txt    139
>
> Can I get a model.Matrix out of that? Or is it not possible to normalize
> that way of hybridization?
>
> Thanks!
>
> Sebastian