[BioC] gcrma vs. rma

[Ben Bolstad]

The two routines vary on how the background adjustment is carried out.
If you view computing expression values as a three stage process

1. Background Correction
2. Normalization
3. Summarization

then in the case of rma()

step 1 is a convolution model
step 2 is quantile normalization
step 3 is a robust multichip model fit using median polish

in the case of gcrma()

step 1 is a based on a model using GC content
step 2 is quantile normalization
step 3 is a robust multichip model fit using median polish

I would not expect the exact same values from both methods.

Thanks,

Ben

On Fri, 2003-12-12 at 09:22, Stan Smiley wrote:
> Sorry if this should have been sent directly to Zhijin Wu, but here goes...
> 
> Is there another resource for learning what the gcrma package is all about?
> Like the theory/rationale behind it and why/how it is better than rma?
> 
> I must be doing something wrong, but my rma expression levels are the same
> as what I got from gcrma.
> 
> Granted I just ran:
> 
> data <- ReadAffy()
> eset.rma <- rma(data)
> eset.gcrma <- gcrma(data)
> 
> But the expression levels generated were identical. I didn't believe it, so
> I'm re-running the analysis just to make sure. It's running now, but since
> gcrma takes so much longer to run, I thought I'd throw this question out
> now.
> 
> I've read the "Textual description of gcrma" pdf file, but it's so cursory
> I'm still left wondering if I just didn't miss the more detailed explanation
> of the package.
> 
> 
> Stan Smiley
> stan.smiley at genetics.utah.edu
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor