[BioC] Applying separate normalization methods to replicate arrays

[Seungwoo Hwang]

I have a conceptual question on normalization of replicate arrays.

Let's say I have several replicate arrays. I wonder whether it is
advisable or not to apply different normalization methods to each of
the replicate arrays. Let me illustrate with the following two
situations. 

(1) By inspecting the raw data, some arrays look as if they need
print-tip lowess while the other arrays look as if global lowess
suffice. Then can I do print-tip lowess to some and global lowess to
others? Or is it advisable to apply print-tip lowess to all replicate
arrays consistently?

(2) By inspecting the raw data, some arrays look as if they need
print-tip scale normalization while the other arrays look as if no such
scaling is needed. Then can I do scaling to some and no scaling to
others? Or it is advisable to apply print-tip scaling to all replicate
arrays consistently?

>From the archive I found a previous post implying consistent
normalization to replicate arrays. 
To quote it: "Normalization together presumes that the conditions were
essentially the same and separate normalization allows them to be
different. When they are the same, then separate normalization will
almost surely be a bit less efficient (in a statistical sense) and when
they are really different joint normalization can be very problematic."

However, if I normalize the replicate arrays with the same method
altogether, wouldn't it be not advisable either because doing so may
remove true biological signals present in the "other" replicate arrays?
For example, in (1), I think that print-tip lowess normalizing the
arrays that seemingly need only global lowess may remove true
biological signals that present in those arrays. Similar arguments may
hold for (2).

In summary, I would like to know which approach is more advisable: (A)
normalizing all replicate arrays using the same method consistently
regardless of differences observed among the replicate arrays, or (B)
using whatever normalization method that best suits each of the
replicate arrays.

Thanks,

Seungwoo

------------------------------------
Seungwoo Hwang, Ph.D.
Senior Research Scientist
Korean Bioinformation Center



      ____________________________________________________________________________________
Never miss a thing.  Make Yahoo your home page.