[BioC] quantile normalization with common standards

[Wolfgang Huber]

Hi Ido

a) yes

b) I don't know a function in Bioconductor that lets you do this easily
for quantiles (it should not be too hard to program this yourself using
interpolation, but I would be concerned about the precision of the
method if your number of control oligos is not very large). Chapter 7 of
this vignette explains how to do it with vsn (which has fewer parameters
to estimate and hence might be more precise):
http://www.bioconductor.org/packages/2.4/bioc/vignettes/vsn/inst/doc/vsn.pdf

Best wishes
 Wolfgang

------------------------------------------------------------------
Wolfgang Huber  EBI/EMBL  Cambridge UK  http://www.ebi.ac.uk/huber


04/11/2008 14:12 Ido M. Tamir scripsit
> Hi,
> 
> suppose I have a high density oligo array. This contains control
> oligos, whose distribution should be the same across experiments (spike in),
> and experimental oligos, whose distribution differs.
> 
> What I did until now was to scale my experimental oligos by the MAD of the 
> control oligos ( a modification of limmas normalizeMedianAbsValues): I 
> calculate a scaling factor for the control oligos and apply this factor onto 
> the experimental oligos of each array.
> 
> like:
> cmed <- log(apply(abs(M[scaleSubset,]), 2, mean, na.rm=TRUE))
> cmed <- exp(cmed - mean(cmed))
> print( paste("scaling factors:", paste( formatC(cmed, digits=2), 
> collapse= " ")  ))
> t(t(M)/cmed)
> 
> Now I would like to go one step further.
> I would like to quantile normalize the control oligos. Then I would like to 
> apply the transformation that was necessary for each array onto the 
> experimental oligos.
> 
> Maybe this makes my intention clearer:
> I don't want to apply the derived averaged distribution onto my experimental 
> oligos - they are not the same (and I anyway know how to do this). I want to 
> do the opposite of that and apply the function that was necessary to 
> transform the control oligos of each array to the averaged quantile 
> distribution onto my experimental oligos.
> Its basically scaling like above, but not only taking MAD as a parameter,
> but the whole difference in the distribution between
> 
> 
> a) does this make sense.
> b) how do I do it.
> 
> thank you very much,
> ido