Hi all, I have a concern about using Aquantile normalization (limma) in my experiment. I hope someone can help me and clarify the issue.... I have several timepoints... in particular one of them, as expected, shows a global, low trascriptional activity (the tissue at this timepoint is close to the "drying stage", so most of the genes are not expressed)... so if I apply Aquantile normalization I am going to modify the channel densities, so that they can overlap. and this is necessary, as far as I understand, because it makes the between-timepoints comparisons possible (my design is unconnected). but, in this way, am I introducing some artifacts in my low-expression timepoints? I mean, I am forcing the channel intensities to have all the same distribution... but which is the assumption of Aquantile ? should the single channel intensities be roughly the same before normalization? thank you ! any help welcome ! Federico

>Date: Fri, 11 Mar 2005 18:41:52 -0800 >From: Federico Scossa <fscossa@pw.usda.gov> >Subject: [BioC] Aquantile normalization & transcriptional activity >To: fscossa@pw.usda.gov >Cc: bioconductor@stat.math.ethz.ch > >Hi all, > >I have a concern about using Aquantile normalization (limma) in my >experiment. I hope someone can help me and clarify the issue.... I have >several timepoints... in particular one of them, as expected, shows a >global, low trascriptional activity (the tissue at this timepoint is close >to the "drying stage", so most of the genes are not expressed)... >so if I apply Aquantile normalization I am going to modify the channel >densities, so that they can overlap. and this is necessary, as far as I >understand, because it makes the between-timepoints comparisons possible >(my design is unconnected). but, in this way, am I introducing some >artifacts in my low-expression timepoints? I mean, I am forcing the >channel intensities to have all the same distribution... but which is the >assumption of Aquantile ? should the single channel intensities be roughly >the same before normalization? You are correct in your suspicion. If one of your samples is expected to show systematically lower expression over the whole genome, then the basic assumptions behind quantile normalization is invalidated. Your options in such a situation are limited. Depending on what is printed on your arrays, you could normalize on a subset of control spots which should have nearly constant expression. You are going to need some sort of boutique normalization. Gordon >thank you ! any help welcome ! > >Federico

yes, the same amount of RNA has been retrotranscibed and applied to all arrays in my experiment. is it still ok to use quantile normalization in this case? thank you for all your help, Federico >is the same amount of mRNA applied to the chip in all cases? if so, it >seems global suppression of mRNA would be compensated for by the use of >more cells, i.e. that normlization would still be OK. > > >----- Original Message ----- From: "Gordon Smyth" <smyth@wehi.edu.au> >To: "Federico Scossa" <fscossa@pw.usda.gov> >Cc: <bioconductor@stat.math.ethz.ch> >Sent: Saturday, March 12, 2005 6:14 AM >Subject: [BioC] Aquantile normalization & transcriptional activity > > >> >>>Date: Fri, 11 Mar 2005 18:41:52 -0800 >>>From: Federico Scossa <fscossa@pw.usda.gov> >>>Subject: [BioC] Aquantile normalization & transcriptional activity >>>To: fscossa@pw.usda.gov >>>Cc: bioconductor@stat.math.ethz.ch >>> >>>Hi all, >>> >>>I have a concern about using Aquantile normalization (limma) in my >>>experiment. I hope someone can help me and clarify the issue.... I have >>>several timepoints... in particular one of them, as expected, shows a >>>global, low trascriptional activity (the tissue at this timepoint is close >>>to the "drying stage", so most of the genes are not expressed)... >>>so if I apply Aquantile normalization I am going to modify the channel >>>densities, so that they can overlap. and this is necessary, as far as I >>>understand, because it makes the between-timepoints comparisons possible >>>(my design is unconnected). but, in this way, am I introducing some >>>artifacts in my low-expression timepoints? I mean, I am forcing the >>>channel intensities to have all the same distribution... but which is the >>>assumption of Aquantile ? should the single channel intensities be roughly >>>the same before normalization? >> >>You are correct in your suspicion. If one of your samples is expected to >>show systematically lower expression over the whole genome, then the >>basic assumptions behind quantile normalization is invalidated. Your >>options in such a situation are limited. Depending on what is printed on >>your arrays, you could normalize on a subset of control spots which >>should have nearly constant expression. You are going to need some sort >>of boutique normalization. >> >>Gordon >> >>_______________________________________________ >>Bioconductor mailing list >>Bioconductor@stat.math.ethz.ch >>https://stat.ethz.ch/mailman/listinfo/bioconductor

>Date: Sat, 12 Mar 2005 12:18:16 -0800 >From: Federico Scossa <fscossa@pw.usda.gov> >Subject: Re: [BioC] Aquantile normalization & transcriptional activity >To: fscossa@pw.usda.gov >Cc: bioconductor@stat.math.ethz.ch > >yes, the same amount of RNA has been retrotranscibed and applied to all >arrays in my experiment. is it still ok to use quantile normalization in >this case? The basic assumption made by quantile normalization is clear, that there is no genome-wide shift in transciptional activity between the RNA targets. This is true for many or most experiments, but only you know enough about your own experiment to judge whether this is reasonable and meaningful in your case. >thank you for all your help, >Federico > > >is the same amount of mRNA applied to the chip in all cases? if so, it > >seems global suppression of mRNA would be compensated for by the use of > >more cells, i.e. that normlization would still be OK. You are citing Tomas Radivoyevitch here, and he may be right, but I worry about what meaning can be ascribed to "differential expression" in such a case. Gordon > >----- Original Message ----- From: "Gordon Smyth" <smyth@wehi.edu.au> > >To: "Federico Scossa" <fscossa@pw.usda.gov> > >Cc: <bioconductor@stat.math.ethz.ch> > >Sent: Saturday, March 12, 2005 6:14 AM > >Subject: [BioC] Aquantile normalization & transcriptional activity > > > >> > >>>Date: Fri, 11 Mar 2005 18:41:52 -0800 > >>>From: Federico Scossa <fscossa@pw.usda.gov> > >>>Subject: [BioC] Aquantile normalization & transcriptional activity > >>>To: fscossa@pw.usda.gov > >>>Cc: bioconductor@stat.math.ethz.ch > >>> > >>>Hi all, > >>> > >>>I have a concern about using Aquantile normalization (limma) in my > >>>experiment. I hope someone can help me and clarify the issue.... I have > >>>several timepoints... in particular one of them, as expected, shows a > >>>global, low trascriptional activity (the tissue at this timepoint is close > >>>to the "drying stage", so most of the genes are not expressed)... > >>>so if I apply Aquantile normalization I am going to modify the channel > >>>densities, so that they can overlap. and this is necessary, as far as I > >>>understand, because it makes the between-timepoints comparisons possible > >>>(my design is unconnected). but, in this way, am I introducing some > >>>artifacts in my low-expression timepoints? I mean, I am forcing the > >>>channel intensities to have all the same distribution... but which is the > >>>assumption of Aquantile ? should the single channel intensities be roughly > >>>the same before normalization? > >> > >>You are correct in your suspicion. If one of your samples is expected to > >>show systematically lower expression over the whole genome, then the > >>basic assumptions behind quantile normalization is invalidated. Your > >>options in such a situation are limited. Depending on what is printed on > >>your arrays, you could normalize on a subset of control spots which > >>should have nearly constant expression. You are going to need some sort > >>of boutique normalization. > >> > >>Gordon