[BioC] Loess normalization and Reference Designs

[michael watson (IAH-C)]

Hi

I have an experiment with a reference design.  My reference is a good
one, in that it lights up every spot brightly, is constant and there is
lots of it.  The result of this, when I look at an M vs A plot, is that
there is a high positive correlation between M and A - as my reference
(Cy3) remains pretty much constant, then as Cy5 increases, so do both M
(log(ratio)) and A(average log(intensity)) (this can be demonstrated by
plotting M vs log(Cy5) and M vs log(Cy3) - the positive correlation
between M and A definitely comes from the Cy5 values)

This is presumably the correct shape for an M vs A plot for a reference
design?  So is it now right to perform loess normalisation?  I expect
the positive correlation between M and A, so is it right to remove it?
If not, then what other normalisation method should I use given that I
want to compare across several arrays all with the same reference?  

Thanks
Mick