[BioC] 2-colour bias AFTER normalising with Limma, labelling QC

[michael watson (IAH-C)]

>But how do people usually QC the labelling and ensure that exactly the
same amount of the 2 samples are mixed?

Matt

As far as I am aware, this is the kind of thing that normalisation
accounts for - different amounts of starting RNA and different labelling
efficiencies of the dyes etc.  However, what these assume is that if
there IS a difference in either of these things, that it is a difference
which is consistent throughout the sample.  We can check one source of
this by looking at spatial effects, but if you can see NO spatial
effects in either Rb, Gb, Rf, Gf or log(ratio) then I'm at a bit of a
loss.

Clearly you have a finger which represents spots that have higher Cy5
than Cy3.  Clearly this finger shows an unnatural relationship with
log(intensity).  You are right to ask - is this a result of unnaturally
high Cy5 or unnaturally low Cy3?  Here's what I would do (if you are
absolutely determined not to bin these results).

1) Filter out the "noise" ie everything that's not the finger.  This
should be pretty easy if you are reasonably accomplished in R - just
filter out all the spots that have an M-value less than 0.2 (an
arbitrary number I have come up with by eye-balling your MvsA plot).

2) Look at these data points - are there any genes or sets of genes
over-represented?  Are any cDNA/oligo subsets over-represented?  Now can
you see if there are any spatial effects or over-represented print-tips?
Randomly pick some of these spots, go back to your plates and do some
PCR or sequencing - do you find what you would expect?  

3) ... Depends on your answers to 2.  Talk to the experimenter.  Make it
clear that any mistakes, errors or differences between the ways
microarrays are treated will give you differential expression of spots
due to systematic error, not due to gene expression.  

... And good luck!  You may also find help on the
microarray-norm at ebi.ac.uk mailing list

Thanks
Mick