Hi Siobhan, I'm not the greatest expert on this but making the DISTRIBUTION look the same is exactly what normalisation (across chips) is meant to do. That is not to say that the EXPRESSION EFFECTS are lost. There are many explanations around but for me the penny dropped best after seeing the explanation on quantile normalisation in the PLIER presentation from affymetrix (requires registration but here's link: https://www.affymetrix.com/support/learning/expression_data/eda_series .affx) . So in short; don't be worried too much by the distribution looking the same whilst the biology is different. I am comparing in vitro vs. orthotopic injection and although in excess of 10% of the genes differ SIGNIFICANTLY by RMA, GC-RMA or PLIER the global expression distributions (i.e. box- plots) look very similar after normalisation. As I don't know why you are doing what you're doing: keep in mind that for LOW expressors any of these global "equalisers" seem to introduce differing degrees of artefactual gene-linkage which is why you should either ignore the low-expression end of the spectrum and why some people still like to use MASS5 when studying pathway-linkage. How much is low and various work-arounds are the subject of plenty publications. No doubt this will keep statisticians of the street for a while yet. Bets regards, Nick N.V. Henriquez, Senior Research Associate Dept. Of Neurodegenerative Diseases Institute of Neurology, UCL, Queen Square House rm 124 Queen Square London WC1N 3BG Tel.? +44 2078373611 ext. 4150 Fax?? +44 2076762157 ------------------------------ Message: 13 Date: Fri, 20 Jun 2008 17:46:09 -0700 From: "Siobhan A. Braybrook" <sabraybrook@ucdavis.edu> Subject: [BioC] Normalization Recommendations- severe biological variation To: bioconductor at stat.math.ethz.ch Message-ID: <6.1.2.0.2.20080620174011.053cab40 at mail.ucdavis.edu> Content-Type: text/plain Hello All! I am hoping that someone might have some suggestions for a normalization method when some of the samples in an experiment are very divergent due to biology, not artifact. I have tried out several (loess, rma, vsn, quantile) but I am worried by how similar the distributions look afterwards. It would be best to use a set of 'housekeeping genes'? The normal ones (rRNA, gapdh, actin, etc) all are biologically different in these treatments too (think dying tissue). Formally we were using mas5 type summarization......but since it isn't the most robust I wanted to try some other methods out. Is the mas5 type of constant normalization really the best for this type of data and I am chasing my tail? Thanks for any advice! Siobhan S. A. Braybrook Graduate Student, Harada Lab Section of Plant Biology University of California, Davis Davis, CA 95616 Ph 530.752.6980 The time is always right, to do what is right. - Martin Luther King, Jr. [[alternative HTML version deleted]] ------------------------------ _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor End of Bioconductor Digest, Vol 64, Issue 21

It really depends on how many genes differ. E.g. if your mechanism turns off expression in a large network, normalization is going to skew your results in the wrong direction. On the other hand, sample preparation which usually results in roughly the same amount of mRNA or total RNA in your samples also skews things in the wrong direction - i.e. in the dying tissue, you are using more tissue to obtain the same quantity of mRNA. --Naomi At 03:56 AM 6/23/2008, Nick Henriquez wrote: >Hi Siobhan, > >I'm not the greatest expert on this but making the DISTRIBUTION look the >same is exactly what normalisation (across chips) is meant to do. That is >not to say that the EXPRESSION EFFECTS are lost. There are many explanations >around but for me the penny dropped best after seeing the explanation on >quantile normalisation in the PLIER presentation from affymetrix (requires >registration but here's link: >https://www.affymetrix.com/support/learning/expression_data/eda_serie s.affx) >. > >So in short; don't be worried too much by the distribution looking the same >whilst the biology is different. I am comparing in vitro vs. orthotopic >injection and although in excess of 10% of the genes differ SIGNIFICANTLY by >RMA, GC-RMA or PLIER the global expression distributions (i.e. box- plots) >look very similar after normalisation. > >As I don't know why you are doing what you're doing: keep in mind that for >LOW expressors any of these global "equalisers" seem to introduce differing >degrees of artefactual gene-linkage which is why you should either ignore >the low-expression end of the spectrum and why some people still like to use >MASS5 when studying pathway-linkage. How much is low and various >work-arounds are the subject of plenty publications. No doubt this will keep >statisticians of the street for a while yet. > >Bets regards, > >Nick > > >N.V. Henriquez, Senior Research Associate >Dept. Of Neurodegenerative Diseases >Institute of Neurology, UCL, >Queen Square House rm 124 >Queen Square >London WC1N 3BG >Tel. +44 2078373611 ext. 4150 >Fax +44 2076762157 > > > > >------------------------------ > >Message: 13 >Date: Fri, 20 Jun 2008 17:46:09 -0700 >From: "Siobhan A. Braybrook" <sabraybrook at="" ucdavis.edu=""> >Subject: [BioC] Normalization Recommendations- severe biological > variation >To: bioconductor at stat.math.ethz.ch >Message-ID: <6.1.2.0.2.20080620174011.053cab40 at mail.ucdavis.edu> >Content-Type: text/plain > >Hello All! > >I am hoping that someone might have some suggestions for a normalization >method when some of the samples in an experiment are very divergent due to >biology, not artifact. >I have tried out several (loess, rma, vsn, quantile) but I am worried by >how similar the distributions look afterwards. >It would be best to use a set of 'housekeeping genes'? The normal ones >(rRNA, gapdh, actin, etc) all are biologically different in these >treatments too (think dying tissue). >Formally we were using mas5 type summarization......but since it isn't the >most robust I wanted to try some other methods out. Is the mas5 type of >constant normalization really the best for this type of data and I am >chasing my tail? > >Thanks for any advice! >Siobhan > >S. A. Braybrook >Graduate Student, Harada Lab >Section of Plant Biology >University of California, Davis >Davis, CA 95616 >Ph 530.752.6980 > >The time is always right, to do what is right. >- Martin Luther King, Jr. > [[alternative HTML version deleted]] > > > >------------------------------ > >_______________________________________________ >Bioconductor mailing list >Bioconductor at stat.math.ethz.ch >https://stat.ethz.ch/mailman/listinfo/bioconductor > > >End of Bioconductor Digest, Vol 64, Issue 21 > >_______________________________________________ >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 Naomi S. Altman 814-865-3791 (voice) Associate Professor Dept. of Statistics 814-863-7114 (fax) Penn State University 814-865-1348 (Statistics) University Park, PA 16802-2111