[BioC] miRNA array normalization - RMA vs Genespring comparison

[andrea.grilli at ior.it]

Dear BioC list,

I'm analyzing Agilent miRNa one color arrays in 3 cell lines using  
AgiMicroRna package (more in detail, it's a time course experiment  
with 2/3 time points, each in replicate for each cell line, for a  
total of 16 samples). I normalized with RMA and used linear models as  
suggested in the manual of the package, getting my "list of d.e. miRs"  
at each time point.
Same data were previously normalized with Genespring and, because of  
the different normalization, this result has poor overlap with mine.
My doubt is not related to the different miRs d.e., but mainly to the  
fact that in my case I have modest modulations (about 1.5-2.5 Fold  
change of absolute values), instead in the second case there are  
changes up to 90 FCs.
You can see an image that can clarify the situation at:
http://www.mediafire.com/i/?r1s3bir4tqs0fmy
Genespring normalization on the left, RMa normalization on the right.

With RMA normalization samples show a clear, more uniform, spread, but  
they are really squeezed on the bottom, so the smaller fold changes.
Could be that RMA normalization had "stretch" too much the data?
Can I support RMA approach in some way to say that could be "better"  
than the other using not only this image (or maybe the other is more  
correct)?

Thanks,
Andrea