[BioC] Suitability of normalizeBetweenArrays? - arrays with very different characteristics

[Gordon Smyth]

Dear John,

You do need to normalize between arrays if you want to do a single 
channel analysis. It is true that normalization is more difficult if 
there are very large differences between samples, but you have to do 
the best you can. I personally use Aquantile in most cases, based on 
unpublished studies in my own group.

Best wishes
Gordon

>Date: Sun, 22 Apr 2007 03:07:55 +0000 (UTC)
>From: John Fowler <fowlerj at science.oregonstate.edu>
>Subject: [BioC] Suitability of normalizeBetweenArrays? - arrays with
>         very    different characteristics
>To: bioconductor at stat.math.ethz.ch
>
>Hello all,
>
>I am looking for advice on whether it is suitable to use 
>normalizeBetweenArrays
>(in limma) in my two-color array experiment.  Secondarily, if it is NOT
>appropriate, would that also preclude doing a single channel analysis of my
>data?  It seems that the User's Guide indicates that some form of normalizing
>between arrays is recommended before doing the single channel analysis.
>
>I have a 3x2 loop design, with three different developmental stages & two
>different genotypes, four replicates each.  I am using spotted long oligo
>arrays, two colors.  Because the two genotypes have very few expression
>differences between them, but two of the developmental stages appear 
>to be VERY
>different, the results on my arrays are also different.  On arrays 
>in which the
>same developmental stage, but different genotypes, are used, the data are
>primarily clustered around M=0, distributed along the A axis.  And, 
>as you would
>expect, when the very different developmental stages of the same genotype are
>used, the amount of variation in M (and presumably in A, as well, although
>that's more difficult to see) is obvious in the plots.
>
>So, is it in-advisable to use normalizeBetweenArrays in this case?  My best
>guess as to the most appropriate method to use would be "Aquantile", but I am
>unsure of that, as well.
>
>I would be happy for suggestions on these questions -
>
>thank you very much,
>John
>
>John Fowler
>Associate Professor
>Oregon State University