[BioC] RMA verse GCRMA, low intensity

Matthew Hannah Hannah at mpimp-golm.mpg.de
Wed Mar 9 17:15:44 CET 2005


Fangxin,

You might want to experiment with eset <- justGCRMA(fast=F) as the fast
version sets signals (or BGs?) that are close to BG to a constant value
(search the lists for a better (correct) definition) which may give some
bias for low signal values. However, you will still have differences in
the BG correction of the 3 methods (MAS,RMA,GCRMA) that will give
differences when signal is close to BG whatever you do. You can easily
see this by plotting the expression values for a single chip for
MAS,RMA,GCRMA,GCRMA(fast=T) against each other.

I was then going to suggest looking for significant differences between
the different timepoints, because much of the low intensity stuff is
non-significant, presumably because it is variable between arrays.
However, you have no replica so this is not possible. Maybe, depending
on what times you have, split the times to form pairs or triplicates of
arrays that you could test against the other arrays. To try and identify
all genes that cycle in potentially complex ways is alot to expect
without replication. If you had replica, most of these 'cycling' genes
that vary between normalizations would probably be eliminated as noise.

BTW - justGCRMA(fast=F) is incredably slow.

HTH,
Matt



More information about the Bioconductor mailing list