Save rma normalization results for later use

0

Entering edit mode

Christian Ruckert ▴ 170

@christian-ruckert-3294

Last seen 4.9 years ago

Germany

I have a bunch of 2000 arrays I want to normalize with rma() from affy package. Then from time to time there will be single arrays to be analyzed together with these 2000. To apply the same normalization procedure to the single arrays later I want to split the rma step in its elements. bg.correct(data, method="rma") As it's array wise I think no problem for the single array. normalize(data, method="quantiles") I think here I need to save the mean values for each row to normalize the single array later with this values (I know it's not totally exact but I think acceptable). In my understanding of the quantile normalization the sorted perfect match values should be exactly the same for every sample, but I got differences. So my questions are: 1. How does the last step look to got exactly the same results as with rma() 2. Why the differences in quantile normalization? 3. Is there a better way to handle this task? Any help would be appreciated, Christian

• 1.0k views

ADD COMMENT • link updated 15.2 years ago by Laurent Gatto 1.6k • written 15.2 years ago by Christian Ruckert ▴ 170

0

Entering edit mode

Wolfgang Huber ★ 13k

@wolfgang-huber-3550

Last seen 17 days ago

EMBL European Molecular Biology Laborat…

Hi Christian, there is some subtlety how quantile normalisation deals with ties (values on one array that are exactly the same), but I think your real problem will be the probset summarisation step, where you need to extract, store, and later apply the probe weights in the RMA model. Henrik Bengtsson recently suggested (on the bioc-devel list): For the purpose of fitting the RMA-style log-additive model, I'd say that Ben [Bolstad]'s robust estimators implemented in preprocessCore are much better (and more flexible, e.g. support weights) than using median polish. See help("rcModelPLM", package="preprocessCore") Best wishes Wolfgang ---------------------------------------------------- Wolfgang Huber, EMBL-EBI, http://www.ebi.ac.uk/huber Ruckert wrote: > I have a bunch of 2000 arrays I want to normalize with rma() from affy > package. Then from time to time there will be single arrays to be > analyzed together with these 2000. To apply the same normalization > procedure to the single arrays later I want to split the rma step in its > elements. > > bg.correct(data, method="rma") > As it's array wise I think no problem for the single array. > > normalize(data, method="quantiles") > I think here I need to save the mean values for each row to normalize > the single array later with this values (I know it's not totally exact > but I think acceptable). In my understanding of the quantile > normalization the sorted perfect match values should be exactly the same > for every sample, but I got differences. > > So my questions are: > 1. How does the last step look to got exactly the same results as with > rma() > 2. Why the differences in quantile normalization? > 3. Is there a better way to handle this task? > > Any help would be appreciated, > > Christian > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor --

ADD COMMENT • link 15.2 years ago Wolfgang Huber ★ 13k

0

Entering edit mode

To try out quickly, the RefPlus package can be an "off-the-shelf" option. L. Wolfgang Huber wrote: > > Hi Christian, > > there is some subtlety how quantile normalisation deals with ties > (values on one array that are exactly the same), but I think your real > problem will be the probset summarisation step, where you need to > extract, store, and later apply the probe weights in the RMA model. > > Henrik Bengtsson recently suggested (on the bioc-devel list): > > For the purpose of fitting the RMA-style log-additive model, I'd say > that Ben [Bolstad]'s robust estimators implemented in preprocessCore are > much > better (and more flexible, e.g. support weights) than using median > polish. See > > help("rcModelPLM", package="preprocessCore") > > > > Best wishes > Wolfgang > > ---------------------------------------------------- > Wolfgang Huber, EMBL-EBI, http://www.ebi.ac.uk/huber > > > > Ruckert wrote: >> I have a bunch of 2000 arrays I want to normalize with rma() from affy >> package. Then from time to time there will be single arrays to be >> analyzed together with these 2000. To apply the same normalization >> procedure to the single arrays later I want to split the rma step in >> its elements. >> >> bg.correct(data, method="rma") >> As it's array wise I think no problem for the single array. >> >> normalize(data, method="quantiles") >> I think here I need to save the mean values for each row to normalize >> the single array later with this values (I know it's not totally exact >> but I think acceptable). In my understanding of the quantile >> normalization the sorted perfect match values should be exactly the >> same for every sample, but I got differences. >> >> So my questions are: >> 1. How does the last step look to got exactly the same results as with >> rma() >> 2. Why the differences in quantile normalization? >> 3. Is there a better way to handle this task? >> >> Any help would be appreciated, >> >> Christian >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor at stat.math.ethz.ch >> https://stat.ethz.ch/mailman/listinfo/bioconductor >> Search the archives: >> http://news.gmane.org/gmane.science.biology.informatics.conductor > > > -- > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor

ADD REPLY • link 15.2 years ago Laurent Gautier ★ 2.3k

0

Entering edit mode

Laurent Gatto 1.6k

@laurent-gatto-5645

Last seen 1 day ago

Belgium

Dear Christian, I think that there are two packages that implement this approach. One is 'RefPlus', available in Bioconductor. There is also fRMA (frozen RMA) from Rafael Irizarry and Matthew McCall, but I don't know if it is available yet. Hope this helps. Laurent On Tuesday 17 February 2009 14:20:55 Christian Ruckert wrote: > I have a bunch of 2000 arrays I want to normalize with rma() from affy > package. Then from time to time there will be single arrays to be > analyzed together with these 2000. To apply the same normalization > procedure to the single arrays later I want to split the rma step in its > elements. > > bg.correct(data, method="rma") > As it's array wise I think no problem for the single array. > > normalize(data, method="quantiles") > I think here I need to save the mean values for each row to normalize > the single array later with this values (I know it's not totally exact > but I think acceptable). In my understanding of the quantile > normalization the sorted perfect match values should be exactly the same > for every sample, but I got differences. > > So my questions are: > 1. How does the last step look to got exactly the same results as with > rma() 2. Why the differences in quantile normalization? > 3. Is there a better way to handle this task? > > Any help would be appreciated, > > Christian > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor

ADD COMMENT • link 15.2 years ago Laurent Gatto 1.6k

Login before adding your answer.