Hello everybody, I am working with dual channel cDNA microarray in a classical profile design: Cy5 different samples Cy3 common reference to identify different classes of samples (unsupervised) or to find the genes associated with our labeles, i.e. desease free survival and so on. I am using LIMMA package to read, preprocessing, normalizing withing arrays and standardizing through the arrays the signals, and QT clustering method (TMEV) for the unsupervised analysis, and LIMMA(i.e. chap 10.5 of LIMMA userguide) or BRB for supervised analysis. I am aware the in such conditions, the homogeneity of reference signals is essential!! so I perform classical plotDensities after standardizing through the arrays (vsn and Aquantile methods): global signals (the densities) are now quite well. To check now the signals reference quality, I converted the MAList into RGList, then I divided Cy3-ref signals of the genes by some global array quantity of Cy3-ref (mean for example) and then I use summary to have some basic statistics. >RG.vAq<- RG.MA(MA.vAglocq) >globalMeans <- apply(RG.vAq$G,2,mean) >len<-dim(RG.vAq$G) >vectorGlobalsMeans<- rep(globalsMeans, len[1]) >arrayGlobalsMeans<-t(array(vectorGlobalsMeans, dim=(len[2],len[1])) >greenSignals<-(RG.vAq$G/arrayGlobalsMeans) Now on greenSignals I have the "reference" intensities of each spot, normalized on the overall array intensities, and so I can measure the variance through all arrays of each genes. So I can eventually filter out which ones have a great variance. Is this procedure correct? There is another procedure to "verify that the gene signals in the reference are valid to confront the expression in the samples?" Thank you in advance Silvano Dr.Silvano Piazza LNCIB, Area Science Park, Padriciano 99 Trieste, ITALY Tel. +39040398992 Fax +39040398990