[BioC] (GC)RMA when there are BIG treatment effects.
Hannah at mpimp-golm.mpg.de
Wed May 19 13:03:52 CEST 2004
I'm looking for opinions on using RMA and GCRMA normalisations
when there are a large amount of changes in expression between
treatments. I remember a comment on this list about this subject
being largely ignored so far, but don't remember a more full
Basically we are going to look at different genotypes, treated and
untreated, with three biological replicates. An initial comparison
comparing 9 genotypes treated versus untreated (no reps yet, just
9 affy chips treated versus 9 untreated chips) by GCRMA, followed
simply by a paired t-test with multtest "BH" multiple testing
correction we see 3000 genes (out of 22000) with p<0.01, and 6000
with p<0.05. Approx 1500 of the p<0.01 have a fold change >2. I've
also seen similar amounts of changes for other experiments such as
day versus night comparisons. This means that the 'majority of
genes not changing' criteria is probably not being met. However,
the number and identity of the changes are biologically meaningful
and so is it justified to continue to use these methods, particulary
in the absence of anything better.
So what do people think/do, surely lots of people are seeing similar
amounts of changes, so how is it being addressed?
More information about the Bioconductor