[BioC] RMA and quantile normalisation

[Francois Collin]

Anne,
There is no reason why the distribution of expression values across a
collection of genes should be normal.  A given gene may have a normal
distribution across samples depending on the selection of samples.  If there
is structure in the sample set (treatment groups, etc), then normality of
errors around the main effects may be a reasonable assumption.  If your
analysis assumes normality for the distribution of expression values for a
given sample across all genes, you may want to compare your results with
those obtained from an analysis that doesn't make this assumption.
Models are fitted to background corrected, normalized probe intensity data
for each probe set separately.  At that level, the distribution of residuals
is not inconsistent with a contaminated normal error distribution for which
a robust estimation procedure as used in RMA makes sense.

-francois


----- Original Message ----- 
From: <Arne.Muller at aventis.com>
To: <bioconductor at stat.math.ethz.ch>
Sent: Wednesday, March 03, 2004 4:16 AM
Subject: [BioC] RMA and quantile normalisation


> Sorry, my last posting was incomplete (slipped over the keyboard ...).
>
> I meant that I haven't explored other methods yet, but since the RMA
values
> are log2, I thought that I'd get something close to a normal distribution.
> Comapred to a normal distribution I get many low intensity probe sets.
>
> The values are generated like this:
>
> eset.rma <- expresso(cel, bgcorrect.method="rma",
>     normalize.method="quantiles", pmcorrect.method="pmonly",
>     summary.method="medianpolish")
>
> then:
> hist(exprs(eset.rma[,10]))
>
> kind regards,
>
> Arne
>
> --
> Arne Muller, Ph.D.
> Toxicogenomics, Aventis Pharma
> arne dot muller domain=aventis com
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>