Hi, Simon, Transformation first or normalization first, it is still a question. I have been using commerce software for years. GeneTraffic and its sister ArrayAssist, for example, typically run normalization first, whether on one color or two color data, then Log transformation, before statistic analysis (such as F-test). Sorin Draghici, in his book of "Data Analysis Tools for DNA Microarrays", suggested workflow for cDNA microarrays in 8 steps (p.337). Log transformation is the step 7, after several normalizations. R package "IlluminaGUI" does quantile normalization without Log transformation first. ... I am new to Illumina and R. Do you have any test data to confirm that your workflow is better than others? Best regards. Longsi Ran, M.D. >From: "Simon Lin" <s-lin2 at="" northwestern.edu=""> >To: "Longsi Ran" <longsiran1126 at="" hotmail.com="">,<dupan at="" northwestern.edu=""> >Subject: Re: [BioC] Workflow question in lumi >Date: Thu, 31 May 2007 10:46:51 -0500 > >I am not sure what you are referring to "the normalization should be done >on linear scales (using not logged raw data)". > >Several microarray textbooks (including Speed; Gentleman; and Baldi) >suggest doing transformation (log or others) first, and normalization >later; because there are distribution assumptions for most normalization >procedures. > >I would be interested to find out a paper or reference doing it in the >reverse order. > >Best regards, > >Simon > >----- Original Message ----- From: "Longsi Ran" <longsiran1126 at="" hotmail.com=""> >To: <dupan at="" northwestern.edu=""> >Cc: <bioconductor at="" stat.math.ethz.ch="">; <s-lin2 at="" northwestern.edu=""> >Sent: Thursday, May 31, 2007 8:29 AM >Subject: [BioC] Workflow question in lumi > > >>Hi, >> >>I noticed that your recommended Illumina workflow in lumi is: 1. VST >>transformation (including Logarithm). 2, RSN or Quantile normalization. >>As most of normalization work on linear expression values, is there a >>particular logic to establish normalizing only logged data? In other >>words, Logarithm first, then Normalization, or vice versa? >> >>I am looking forward to hearing from you soon. >> >>Thanks! >> >>Longsi Ran M.D >> >>_________________________________________________________________ >>Windows Live Hotmail with drag and drop, you can easily move and organize >>your mail in one simple step. Get it today! >>www.newhotmail.ca?icid=WLHMENCA153 >> > > _________________________________________________________________ Windows Live Hotmail with drag and drop, you can easily move and organize www.newhotmail.ca?icid=WLHMENCA153

Hi Longsi, If you take a look of the intensity distribution of the raw data, you can find the majority of the data is very low expresse while some probes are very highly expressed. The log2 or vst (variance stabilizing transform) transform is to suppress this dynamic range of intensity and make the normalization, especially the curve-fitting related normalization, easier to perform. If the normalization algorithm can deal with the raw data (with wide dynamic range of intensity) well, it should have no big difference of the order. Pan On 6/1/07 9:19 AM, "Longsi Ran" <longsiran1126 at="" hotmail.com=""> wrote: > Hi, Simon, > > Transformation first or normalization first, it is still a question. > > I have been using commerce software for years. GeneTraffic and its sister > ArrayAssist, for example, typically run normalization first, whether on one > color or two color data, then Log transformation, before statistic analysis > (such as F-test). > > Sorin Draghici, in his book of "Data Analysis Tools for DNA Microarrays", > suggested workflow for cDNA microarrays in 8 steps (p.337). Log > transformation is the step 7, after several normalizations. > > R package "IlluminaGUI" does quantile normalization without Log > transformation first. > > ... > > I am new to Illumina and R. Do you have any test data to confirm that your > workflow is better than others? > > Best regards. > > Longsi Ran, M.D. > > > > >> From: "Simon Lin" <s-lin2 at="" northwestern.edu=""> >> To: "Longsi Ran" <longsiran1126 at="" hotmail.com="">,<dupan at="" northwestern.edu=""> >> Subject: Re: [BioC] Workflow question in lumi >> Date: Thu, 31 May 2007 10:46:51 -0500 >> >> I am not sure what you are referring to "the normalization should be done >> on linear scales (using not logged raw data)". >> >> Several microarray textbooks (including Speed; Gentleman; and Baldi) >> suggest doing transformation (log or others) first, and normalization >> later; because there are distribution assumptions for most normalization >> procedures. >> >> I would be interested to find out a paper or reference doing it in the >> reverse order. >> >> Best regards, >> >> Simon >> >> ----- Original Message ----- From: "Longsi Ran" <longsiran1126 at="" hotmail.com=""> >> To: <dupan at="" northwestern.edu=""> >> Cc: <bioconductor at="" stat.math.ethz.ch="">; <s-lin2 at="" northwestern.edu=""> >> Sent: Thursday, May 31, 2007 8:29 AM >> Subject: [BioC] Workflow question in lumi >> >> >>> Hi, >>> >>> I noticed that your recommended Illumina workflow in lumi is: 1. VST >>> transformation (including Logarithm). 2, RSN or Quantile normalization. >>> As most of normalization work on linear expression values, is there a >>> particular logic to establish normalizing only logged data? In other >>> words, Logarithm first, then Normalization, or vice versa? >>> >>> I am looking forward to hearing from you soon. >>> >>> Thanks! >>> >>> Longsi Ran M.D >>> >>> _________________________________________________________________ >>> Windows Live Hotmail with drag and drop, you can easily move and organize >>> your mail in one simple step. Get it today! >>> www.newhotmail.ca?icid=WLHMENCA153 >>> >> >> > > _________________________________________________________________ > Windows Live Hotmail with drag and drop, you can easily move and organize > your mail in one simple step. Get it today! > www.newhotmail.ca?icid=WLHMENCA153 >