Hi, Firstly, GCRMA rocks and many thanks to all those involved in producing it. However, GCRMA is painfully slow. I understand there are plans for a justGCRMA. Will the fast version have options for normalisation other than the default quantiles? QN is good for merging replicates, which should all have the same distribution. However, it introduces artifacts when comparing chips from different conditions which are observed to have different distributions. Is it possible to have a quick version for just the background calculation? And possibly another quick version for the median polish expression summary? But I think the normalisation choice should not be hardcoded. Or is this heresy? Thanks again, Harry -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~ Dr Andrew Harrison Tel: 44 (0) 207 679 3890 Biomolecular Structure and Modelling Unit Fax: 44 (0) 207 679 7193 Biochemistry and Molecular Biology Dept. University College London Gower Street Email: harry@biochem.ucl.ac.uk London, WC1E 6BT, UK http://www.biochem.ucl.ac.uk/~harry ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~

justGCRMA is already in the unstable devel version. the internal function gcrma.engine permits you to do just background correction. you can then use expresso to choose whatever other option you want. its not easy to use so we will look into creating an easy to use wrapper for it. On Fri, 19 Mar 2004, Andrew Harrison wrote: > Hi, > > Firstly, GCRMA rocks and many thanks to all those > involved in producing it. > > However, GCRMA is painfully slow. I understand there are plans > for a justGCRMA. Will the fast version have options for normalisation > other than the default quantiles? QN is good for merging replicates, > which should all have the same distribution. However, it introduces > artifacts when comparing chips from different conditions which are > observed to have different distributions. > > Is it possible to have a quick version for just the > background calculation? And possibly another quick version for > the median polish expression summary? But I think the normalisation > choice should not be hardcoded. > > Or is this heresy? > > Thanks again, > Harry > >

> Will the fast version have options for normalisation > other than the default quantiles? QN is good for merging replicates, > which should all have the same distribution. However, it introduces > artifacts when comparing chips from different conditions which are > observed to have different distributions. Almost every sophisticated normalization method would introduce this same "artifact" to one degree or another. However, what I have typically observed is that great differences in distribution between conditions are more often then not due to technical differences (often confounded with the conditions since people typically do not seem to randomize).

Hi, For affymetrix the stringest differences occure during to the RNA extraction and purification steps - at least that's my experience. I'm trying to analyze 3 dataset that "should" be the same (same tratment) but have been generated with slightely different purification kits. The data distributions (visually inspected kernel density plots) are quite different. I guess 'vsn' might be a better choice for when there are strong differences in the distributions since it takes into account the variance effects, e.g. one experimental batch might have higher intensities due to a better amplification. Anyway, it's just a suggestions. Might be realy good if different normalisation methods could be passed to gcrma or justRMA. regards, Arne -- Arne Muller, Ph.D. Toxicogenomics, Aventis Pharma arne dot muller domain=aventis com > -----Original Message----- > From: bioconductor-bounces@stat.math.ethz.ch > [mailto:bioconductor-bounces@stat.math.ethz.ch]On Behalf Of > Ben Bolstad > Sent: 19 March 2004 17:37 > To: Andrew Harrison > Cc: bioconductor@stat.math.ethz.ch > Subject: Re: [BioC] justGCRMA normalisation options? > > > > Will the fast version have options for normalisation > > other than the default quantiles? QN is good for merging replicates, > > which should all have the same distribution. However, it introduces > > artifacts when comparing chips from different conditions which are > > observed to have different distributions. > > Almost every sophisticated normalization method would > introduce this same > "artifact" to one degree or another. However, what I have typically > observed is that great differences in distribution between > conditions are > more often then not due to technical differences (often > confounded with > the conditions since people typically do not seem to randomize). > > _______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor >

Also note that justGCRMA will likely not be faster than gcrma, and the devel versions of both are much faster due to the introduction of some C code (thanks to Jeff Gentry). If you have the memory to run gcrma, and are adept at hacking code, you could just modify gcrma.engine() to return the background corrected AffyBatch and then feed this to expresso. Because of the way justGCRMA works, it would take a bit more work to change how the data are normalized, so if you have the RAM, gcrma is probably the way to go for now. Jim James W. MacDonald Affymetrix and cDNA Microarray Core University of Michigan Cancer Center 1500 E. Medical Center Drive 7410 CCGC Ann Arbor MI 48109 734-647-5623 >>> "Rafael A. Irizarry" <ririzarr@jhsph.edu> 03/19/04 11:20AM >>> justGCRMA is already in the unstable devel version. the internal function gcrma.engine permits you to do just background correction. you can then use expresso to choose whatever other option you want. its not easy to use so we will look into creating an easy to use wrapper for it. On Fri, 19 Mar 2004, Andrew Harrison wrote: > Hi, > > Firstly, GCRMA rocks and many thanks to all those > involved in producing it. > > However, GCRMA is painfully slow. I understand there are plans > for a justGCRMA. Will the fast version have options for normalisation > other than the default quantiles? QN is good for merging replicates, > which should all have the same distribution. However, it introduces > artifacts when comparing chips from different conditions which are > observed to have different distributions. > > Is it possible to have a quick version for just the > background calculation? And possibly another quick version for > the median polish expression summary? But I think the normalisation > choice should not be hardcoded. > > Or is this heresy? > > Thanks again, > Harry > > _______________________________________________ Bioconductor mailing list Bioconductor@stat.math.ethz.ch https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor