Dear listers, I am new to the analysis of 2-dye arrays, so please bear with me! I have a conceptual question on MA plots and dye swaps. Assume you have 2 arrays, on which Control and Treatment are measured + its corresponding dye swap. So: array Cy3 Cy5 1 Con Treatment 2 Treatment Con After normalizing these 2 arrays (e.g. with VSN from within Limma) I plot the 2 MA plots (based on the normalized data). I would expect that (in theory) MA plot1 would be the mirrored version of MA plot2. In other words, normalization per se does NOT take into account dye swaps; this is only subsequently done in the design matrix when using e.g. limma. Q: Is this correct? I am asking because in my normalized dataset with dye swaps (15 arrays total) I do NOT seem to see these mirrored MA plots when comparing the respective dye-swapped arrays. Thanks, Guido Code: library("limma") targets <- readTargets("targets_corrected.txt", row.names="Name") RG <- read.maimages(targets$FileName, source="agilent") MA.vsn <- normalizeBetweenArrays(RG, method="vsn") plotMA3by2(MA.vsn) ------------------------------------------------ Guido Hooiveld, PhD Nutrition, Metabolism & Genomics Group Division of Human Nutrition Wageningen University Biotechnion, Bomenweg 2 NL-6703 HD Wageningen the Netherlands tel: (+)31 317 485788 fax: (+)31 317 483342 internet: http://nutrigene.4t.com <http: nutrigene.4t.com=""/> email: guido.hooiveld@wur.nl [[alternative HTML version deleted]]

Hi Guido! Additionally to what Wolfgang already wrote you might have a look at a related thread on the mailing list, that was discussed just recently (https://stat.ethz.ch/pipermail/bioconductor/2010-January/thread.html# 31254) . What you call "mirrored version" of the MA-Plot, corresponds to the negative correlation of the M-values discussed in that thread. One reason why you tend not to observe this negative correlation/mirrored version is that the normalization (loess,vsn,whatever) removes an OVERALL dye effect, which is also allowed to be intensity depending for the standard methods, BUT: the dye effect can also have a gene-specific component, i.e. two genes with similar intensities might have very different dye-effects for example because of their difference in GC content or other properties. This is not addressed by the standard methods, which means your MA- Plots will not look as "mirrored" as you think. You can account for the gene-specific dye effect though by including that effect in your limma model (cf. the limma tutorial). Also (and this is partly by covered by Wolfgang's reply) you will only observe a mirror effect for genes which are truly biologically differential expressed. So if there is not much differential expression, there won't be much mirroring either. Best Wishes Claus > -----Original Message----- > From: bioconductor-bounces at stat.math.ethz.ch [mailto:bioconductor- > bounces at stat.math.ethz.ch] On Behalf Of Hooiveld, Guido > Sent: 28 January 2010 14:14 > To: bioconductor at stat.math.ethz.ch > Subject: [BioC] MA plots + dye swap > > > Dear listers, > I am new to the analysis of 2-dye arrays, so please bear with me! > I have a conceptual question on MA plots and dye swaps. > > Assume you have 2 arrays, on which Control and Treatment are measured + > its corresponding dye swap. So: > array Cy3 Cy5 > 1 Con Treatment > 2 Treatment Con > > After normalizing these 2 arrays (e.g. with VSN from within Limma) I > plot the 2 MA plots (based on the normalized data). I would expect that > (in theory) MA plot1 would be the mirrored version of MA plot2. In other > words, normalization per se does NOT take into account dye swaps; this > is only subsequently done in the design matrix when using e.g. limma. > Q: Is this correct? I am asking because in my normalized dataset with > dye swaps (15 arrays total) I do NOT seem to see these mirrored MA plots > when comparing the respective dye-swapped arrays. > > Thanks, > Guido > > Code: > library("limma") > targets <- readTargets("targets_corrected.txt", row.names="Name") > RG <- read.maimages(targets$FileName, source="agilent") > MA.vsn <- normalizeBetweenArrays(RG, method="vsn") > plotMA3by2(MA.vsn) > > > > ------------------------------------------------ > Guido Hooiveld, PhD > Nutrition, Metabolism & Genomics Group > Division of Human Nutrition > Wageningen University > Biotechnion, Bomenweg 2 > NL-6703 HD Wageningen > the Netherlands > tel: (+)31 317 485788 > fax: (+)31 317 483342 > internet: http://nutrigene.4t.com <http: nutrigene.4t.com=""/> > email: guido.hooiveld at wur.nl > > > > [[alternative HTML version deleted]] > > _______________________________________________ > 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

Hi Guido, normalisation methods would indeed not know about your experimental design (including the dye swapping). Now, just a request for clarification - what do you mean by "mirrored"? The plots will not be exactly mirrored since they are from different arrays. You would expect to see biological effects that are present in both arrays 'mirrored' w.r.t. the M=0 axis. However, there are also technical effects (either random, or associated with dye), which would not be 'mirrored'. So, are you sure that the biological signal you are expecting is the main contribution to the picture you see in the plots? It would probably be helpful to select a handful of genes that are important to you, and mark them specially (by color or plot symbol) in the plots. Best wishes Wolfgang Il giorno Jan 28, 2010, alle ore 3:14 PM, Hooiveld, Guido ha scritto: Dear listers, I am new to the analysis of 2-dye arrays, so please bear with me! I have a conceptual question on MA plots and dye swaps. Assume you have 2 arrays, on which Control and Treatment are measured + its corresponding dye swap. So: array Cy3 Cy5 1 Con Treatment 2 Treatment Con After normalizing these 2 arrays (e.g. with VSN from within Limma) I plot the 2 MA plots (based on the normalized data). I would expect that (in theory) MA plot1 would be the mirrored version of MA plot2. In other words, normalization per se does NOT take into account dye swaps; this is only subsequently done in the design matrix when using e.g. limma. Q: Is this correct? I am asking because in my normalized dataset with dye swaps (15 arrays total) I do NOT seem to see these mirrored MA plots when comparing the respective dye-swapped arrays. Thanks, Guido Code: library("limma") targets <- readTargets("targets_corrected.txt", row.names="Name") RG <- read.maimages(targets$FileName, source="agilent") MA.vsn <- normalizeBetweenArrays(RG, method="vsn") plotMA3by2(MA.vsn) ------------------------------------------------ Guido Hooiveld, PhD Nutrition, Metabolism & Genomics Group Division of Human Nutrition Wageningen University Biotechnion, Bomenweg 2 NL-6703 HD Wageningen the Netherlands tel: (+)31 317 485788 fax: (+)31 317 483342 internet: http://nutrigene.4t.com <http: nutrigene.4t.com=""/> email: guido.hooiveld at wur.nl [[alternative HTML version deleted]] _______________________________________________ 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