Hi all, I'm picking up on an older thread (https://stat.ethz.ch/pipermail/bioconductor/2006-August/013881.html ) which poses the following question: What are the most appropriate methods for normalization in the face of large numbers of differentially expressed genes? Aside from this 2006 thread, I've found only one publication on the topic (http://www.ncbi.nlm.nih.gov/e ntrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=190407 42 ) but the authors, unfortunately for me, have built their software using Matlab. My data consists of Agilent arrays on eight (very different) tissues hybridized against a common reference sample. In six of the hybridization, Cy3 is the reference and Cy5 is the tissue in question. In two of the arrays, the situation is reversed. I thought I would throw open the doors again with the following questions: 1) In the three years since aforementioned post, are there better methods that I could look into for normalizing in the face of large gene effects? 2) How "robust" are loess-based normalization methods? I've been using the limma package for within array normalization using global loess (no print tips on Agilent arrays), but its frankly hard to evaluate the consequences given the relatively few numbers of spike- ins. 3) Ye olde background correction issue: Because the background does not appear all that uniform on my slides, I would like to do a local background correction. But what to do about those damn negative numbers? My current thoughts is to carry out a background subtraction using "min" in limma so that negative values are set to be equal to the smallest positive value on the array. But perhaps it is better to set the small values = 1 to avoid issues in the log transformation? I'm also concerned about an effect of "min" subtraction that can be seen in the following density plots: Before subtraction: http://www.duke.edu/~dag23/BioCQs/densityPlot_RG_originalData.pdf After subtraction: http://www.duke.edu/~dag23/BioCQs/densityPlot_RG_minSubtracted.pdf These later artifacts appear to lead to an unfortunate pattern after loess normalization (http://www.duke.edu/~dag23/BioCQs/densityPlot_MA_minSubtracted.pdf ) Its a great many questions, I know, but any insight would be greatly appreciated. I suspect that these issues comes up more frequently than I realize, so I apologize if I've missed an earlier thread. Best wishes, David [[alternative HTML version deleted]]