> Date: Wed, 3 Aug 2005 08:27:13 US/Arizona > From: scholz at Ag.arizona.edu > Subject: [BioC] loop design question > To: bioconductor at stat.math.ethz.ch > > Hello, > > I'm new to limma and have been plowing through the excellent users guide, where > I've reached a sticking point. Would someone have the saintly patience to > explain to a small mind how the design matrix for the direct design example > works? I think I'm missing something quite fundamental in that I was under the > impression that the numbers 1 and -1 represented the red and green dyes, > respectively, but if that is true, I have no idea what these numbers mean in > columns headed "CD8-CD4" and "DN-CD4". In fact, I don't really understand what > these "subtracted" column headers mean at all, either in the design matrix or > the contrast matrix. I'm planning a loop design experiment and this appears to > be an essential point to grasp. Thanks in advance of your answer. The headers mean that the coefficients represent the comparisons CD8 vs CD4 and DN vs CD4 respectively. As the text explains, the other two treatments are compared back to CD4. The easiest way to analysis a direct design is to choose one of the treatments to compare back to in this way, i.e., to stand in as a virtual common reference. The design matrix for the CD example could have been computed using design < modelMatrix(targets, ref="CD4") Just use the modelMatrix() function, look at the 1's and -1's, just understand what the interpretatation of the columns is in terms of treatment comparisons. Gordon > Matt Scholz > Research Specialist > Department of Plant Science > University of Arizona > scholz at ag.arizona.edu > > --------------------------------------------- > College of Agriculture and Life Sciences Web Mail. > http://ag.arizona.edu

> Date: Mon, 8 Aug 2005 10:38:10 +0200 > From: Matja? Hren <matjaz.hren at="" nib.si=""> > Subject: Re: [BioC] loop design question > To: <smyth at="" wehi.edu.au="">, <scholz at="" ag.arizona.edu=""> > Cc: bioconductor at stat.math.ethz.ch > > Hello, > > This short debate encouraged me to ask another question regarding direct microarray design and > limma's design matrix. > > We compare response of plants to viral infection an several varieties of the sam plant species > (potato). We don't use a common reference but we hybridise on each array infected (V) and healthy > plant (M - mock infected) - for lets say 2 cultivars ( indexes "s", "i"). There are 3 biological > repetitions including dyswaps - as seen in the targets file (below). > > My question is if the design matrix we created (down below) is OK? In the "CD8 vs CD4 and DN vs > CD4" case described in Limma User's Guide and discussed below the CD4 was used as a virtual common > reference. Do you have any suggestions what could be used as a "virtual common reference" in our > case or it doesn't matter what we choose for it? Well, yes, you could analyse the experiment in principle with any choice of virtual common reference. But your experiment is hardly a "loop design". Your experiment has a definite special structure, which presumably arises from the fact that you are interested in the effect of infection on each of the varieties. A sensible analysis should reflect this structure and should return quantities that are of interest to you. There is no way for a program like limma to give you canned analyses for every possible scientific situation. The package is rather designed to give you the flexibility to tailor the analysis to your experiment. You experiment has been designed to estimate the effect of infection on each variety. You should design the analysis to do the same thing. You could do this by design <- cbind(I=c(-1,1,-1,0,0,0), S=c(0,0,0,-1,1,1)) In your linear model, the first coefficient will now be the effect of V over M or variety "i", while the second coefficient will be the effect of V over M for variety "s". You could use a contrast to test whether the two coefficients are different, i.e., to compare varieties. This approach extends easily to any number of varieties. Gordon > Could we also include additional microarrays with the same experimental layout but done on two > additional varietis of the plants (lets say indexes "c" and "n")in the experiment? And would then > the logics of the design matrix construction still stay the same - see below (pick one virtual > common reference)? Or would it be better if we used ANOVA to compare varieties? > > >> targets<-readTargets("C:/Uporabniki/RRR/phenoDataIS.txt") >> targets > fileName slide.number variety Cy5 Cy3 > 1 I_30m_1_1042.txt 12901042 I Mi Vi > 2 I_30m_2_1109.txt 12901109 I Vi Mi > 3 I_30m_3_1040.txt 12901040 I Mi Vi > 4 S_30m_1_1103.txt 12901103 S Ms Vs > 5 S_30m_2_1039.txt 12901039 S Vs Ms > 6 S_30m_3_1106.txt 12901106 S Vs Ms > >> design <- modelMatrix(targets, ref="Mi") > Found unique target names: > Mi Ms Vi Vs > >> design > Ms Vi Vs > 1 0 -1 0 > 2 0 1 0 > 3 0 -1 0 > 4 1 0 -1 > 5 -1 0 1 > 6 -1 0 1 > > And now for the four varieties (reference would be "Mi"): > > targets<-readTargets("C:/Uporabniki/RRR/phenoDataISCN.txt") > >> design <- modelMatrix(targets, ref="Mi") > Found unique target names: > Mc Mi Mn Ms Vc Vi Vn Vs > >> design > Mc Mn Ms Vc Vi Vn Vs > 1 0 0 0 0 -1 0 0 > 2 0 0 0 0 1 0 0 > 3 0 0 0 0 -1 0 0 > 4 0 0 1 0 0 0 -1 > 5 0 0 -1 0 0 0 1 > 6 0 0 -1 0 0 0 1 > 7 1 0 0 -1 0 0 0 > 8 -1 0 0 1 0 0 0 > 9 1 0 0 -1 0 0 0 > 10 0 1 0 0 0 -1 0 > 11 0 -1 0 0 0 1 0 > 12 0 -1 0 0 0 1 0 > > Thank you for any replies, > > Matjaz > > P.S. I Use RGUI 2.1.1 on WindowsXP and limma package. > > -------------------------------------------------------------------- ------------ > Matjaz Hren > National Institute of Biology > Dept. of Plant Physiology and Biotechnology > Vecna pot 111 > 1000 Ljubljana > SLOVENIA > > www.nib.si > phone: +386 1 423 33 88 > fax: +386 1 257 38 50 > > -------------------------------------------------------------------- ------------