Thanks Gordon, I mixed up as a result of what I read in "Introductory statistic with R" where there is an example of (section 10.2) x<-runif(20) y<-2*x+rnorm(20,0,0.3) summary(lm(y~x)) gives insignificant intercept so there is a suggestion to run summary(lm(y~x-1)) but when I run once model.matrix(y~x) and once model.matrix(y~x-1) I saw my mistake exactly as you said. Ron >>> "Gordon K Smyth" <smyth at="" wehi.edu.au=""> 03/01/06 11:46 PM >>> > Date: Tue, 28 Feb 2006 15:02:32 +0200 > From: "Ron Ophir" <ron.ophir at="" weizmann.ac.il=""> > Subject: [BioC] Another question about dye swap > To: <bioconductor at="" stat.math.ethz.ch=""> > Message-ID: <s404661c.050 at="" wisemail.weizmann.ac.il=""> > Content-Type: text/plain; charset=US-ASCII > > Hi, >>From limma user guide section 8.1.2 Simple comparisons -> Dye swaps it > is clear that for the following experimental design: > FileName Cy3 Cy5 > File1 wt mu > File2 mu wt > File3 wt mu > File4 mu wt > *(four replicates of two groups (wt , mu) of which two replicates in > each group is labeled by one color(red) the other two is labeled by > another color (green)) > one can estimate the Dye effect by defining that effect as the > intercept : > design <- cbind(DyeEffect=1,MUvsWT=c(1,-1,1,-1)) > fit <- lmFit(MA, design) > fit <- eBayes(fit) > > Let's say that one finds that the Dye effect (for all genes) is > statisticaly insignificant. Would it be correct than to force the model > to go through the intercept by > > design <- cbind(DyeEffect=-1,MUvsWT=c(1,-1,1,-1)) > > Will I get a better estimation of the Mu to wt ratio? > > Thanks, > Ron No. The code you give does not force the model through the intercept and would give erroneous estimates. Simply omitting the DyeEffect column does what you want. Gordon

Hi, Is it possible to include the dye effect for those genes for which it's found to be significant and drop it for the others? I guess I could separate the two groups of genes and define a different design matrix for each group, but I'm wondering if there's also a way to produce a "dynamic" kind of design matrix... Thanks, Elena >>> "Gordon K Smyth" <smyth at="" wehi.edu.au=""> 03/01/06 11:46 PM >>> > Date: Tue, 28 Feb 2006 15:02:32 +0200 > From: "Ron Ophir" <ron.ophir at="" weizmann.ac.il=""> > Subject: [BioC] Another question about dye swap > To: <bioconductor at="" stat.math.ethz.ch=""> > Message-ID: <s404661c.050 at="" wisemail.weizmann.ac.il=""> > Content-Type: text/plain; charset=US-ASCII > > Hi, >>From limma user guide section 8.1.2 Simple comparisons -> Dye swaps it > is clear that for the following experimental design: > FileName Cy3 Cy5 > File1 wt mu > File2 mu wt > File3 wt mu > File4 mu wt > *(four replicates of two groups (wt , mu) of which two replicates in > each group is labeled by one color(red) the other two is labeled by > another color (green)) > one can estimate the Dye effect by defining that effect as the > intercept : > design <- cbind(DyeEffect=1,MUvsWT=c(1,-1,1,-1)) > fit <- lmFit(MA, design) > fit <- eBayes(fit) > > Let's say that one finds that the Dye effect (for all genes) is > statisticaly insignificant. Would it be correct than to force the model > to go through the intercept by > > design <- cbind(DyeEffect=-1,MUvsWT=c(1,-1,1,-1)) > > Will I get a better estimation of the Mu to wt ratio? > > Thanks, > Ron No. The code you give does not force the model through the intercept and would give erroneous estimates. Simply omitting the DyeEffect column does what you want. Gordon _______________________________________________ Bioconductor mailing list Bioconductor at stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/bioconductor "EMF <coh.org>" made the following annotations. ---------------------------------------------------------------------- -------- SECURITY/CONFIDENTIALITY WARNING: This message and any atta...{{dropped}}