[BioC] MAS, RMA or DChip. Which is better?
narinder.singh
narinder.singh at diagenic.com
Mon Sep 27 13:08:50 CEST 2004
Hello,
Although I have been using and analyzing gene expresion data for some time, I
have recently (new job)starting looking at Affymetrix data. From what I have
read MAS5.0, RMA and dChip are among the most used methods for "pre-
normalizing" or summarizing the probe-level data. Reading the relevant
literature, most of the comparisons have been made on designed data aimed at
studying specific characteristics of the metods, e.g bias-variance tradeoff
etc.
I have been handed over a data set based on a 2^2 experimental design with 3
reps. per level giving a total of 12 samples measured on Affymetrix arrays,
mouse-430A. Now, my question (problems) are as following.
How should the data be normalized, i.e. local/group-wise normalization, or
global normalization.
The information extracted depends upon the choice of the normalization.
MAS5.0 does not do a group-wise normalization, so RMA and dhip are the two
methods available when doing local-normalization.
Global normalization gives no difference between the four groups (as stuied
on a scores plot obtained using PCA ) irrespective of the method employed,
while local normalization leads to quite distinct results depending upon the
choice of the method employed.
Using RMA (local normal.) one can separate clearly between the four groups on
the first-two PCs, while dChip clusters the groups in a totally different
manner. Thus the interpretation of the results depend upon the choice between
RMA and dChip. This is a major problem.
Using other platforms, the results are affected by the choice of the method
employed to cluster the data, while in Affymetrix the result is not based on
the choice of the method of abalysis, but instead on the "pre-normalization"
step. Biology should and cannot change based on how the data are summarized.
I would appreciate if somebody can point point me to a site or an article
where this problem has been discussed before.
Thanks for your time.
Narinder Singh Sahni, PhD
Rikshospitalet University hospital
Centre for Occupational and Environmental Medicine
0027 Oslo
Norway
Phone (Dept.): +47 - 23073650
--
Open WebMail Project (http://openwebmail.org)
More information about the Bioconductor
mailing list