[BioC] normalizing affy chips with vsn

Oliver Hartmann hartmann@mailer.uni-marburg.de
Wed, 15 Jan 2003 15:20:30 +0100


Dear list members,

I wanted to compare rma and vsn for Affy data as I ran into a data set 
where rma could be improved (the t-statistic was still intensity 
dependent). Everything except the normalization method from rma I wanted 
to stay the same to see if vsn outperforms quantile normalization.
I posted a question on how to use expresso() with vsn befor and thanks 
to a thorough discussion with Wolfgang Huber I managed to put up a 
script. I had to change the median polish function as it log transforms 
the data in the original affy package. This is the script that performed 
well (results very similar to rma) in the Affymetrix latin square data set:

generateExprVal.method.medianpolish2 <- function(probes, ...)
   medianpolish2(probes, ...)
medianpolish2 <- function(x, ...){
   tmp <- medpolish(x, trace.iter=FALSE, ...)
   ##rough estimate
   sigma <- 1.483*median(abs(as.vector(tmp$residuals)))/sqrt(nrow(x))
   list(exprs=tmp$overall + tmp$col,se.exprs=rep(sigma, ncol(x)))
}
generateExprSet.methods <- c(generateExprSet.methods, "medianpolish2")
express.summary.stat.methods <- c(express.summary.stat.methods, 
"medianpolish2")
normalize.AffyBatch.methods <- c(normalize.AffyBatch.methods,"vsn")
es5 = expresso(data,
         pmcorrect.method = "pmonly",
         bgcorrect.method = "rma",
         normalize.method = "vsn",
         summary.method   = "medianpolish2")

Of course, there might be other summary or background correction methods 
than these, but as the results are very close to rma, there won't be 
much room for improvement. So, if you ask me, this would be the script 
to use for affy chips with vsn normalization with the current 
bioconductor version. Any comments? Oh, in case if you are wondering: I 
couldn't try vsn with the data set I mentioned at the beginning - it's 
to big to calculate vsn on my PC, but as soon as I get my new one I'll 
try...

Details of the comparisson of VSN to RMA and MAS can be found at my web 
page: http://staff-www.uni-marburg.de/~hartmann/

Thanks a lot to Wolfgang and all those that answered, too.

	-oli-
-- 
Oliver Hartmann, Institute of Medical Biometry and Epidemiology
Philipps-University Marburg, Bunsenstr. 3, D-35037 Marburg
phone +49(0)6421 28 66514, fax +49(0)6421 28 68921