[BioC] 2-colour cDNA data quality (+different species)

[Matthew Hannah]

Hi,

I've just been looking at some data where a wild relative and model
plant have been hybridised to an array designed for the model plant.
There are only 2 chips - 2 biological replicates with dye swap. I
normalised the data with limma as below -

RG.nb <- backgroundCorrect(RG, method="none") 
MA.nb <- normalizeWithinArrays(RG.nb)
MA.nb.s <- normalizeBetweenArrays(MA.nb)
rg <- RG.MA(MA.nb.s)
plot(rg$R[,1], rg$G[,2])
abline(0,1)

The plot (attached, but I guess it will be scrubbed?) has the majority
of the genes on the line, but there are a large amount of genes all up
regulated (a diffuse 'finger' coming out of the side of the scatter
plot). The dye swap shows exactly the same (for the biological rep).
Even though these are two independent labellings, and the same amount of
labelled cDNA was used, is this likely to be due to different cDNA
amounts or quality. Has anyone else had similar experiences, particulary
when comparing different species?

If this is the case, then why do the majority of genes show a good
correlation? Also in more general terms I've been looking at the
correlations between replicates (R and G values, not ratios) in other
experiments, and whilst in most cases the correlation is good (nice 0,1
scatterplot) in some cases its very poor (think explosion from the
bottom left of the plot!). Can anyone offer any experience or
suggestions on how to identify whether the Biological sample, RNA,
labelling or slide may be at fault.

Thanks,

Matt

P.S - sorry if this doesn't text wrap, our email has been 'upgraded'.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: RvsG.png
Type: image/png
Size: 14275 bytes
Desc: RvsG.png
Url : https://www.stat.math.ethz.ch/pipermail/bioconductor/attachments/20040727/5bb2b4ea/RvsG.png