Dear list, I am working with Affymetrix microarrays and looking for differential gene expression using limma. We repeated the same microarray experiment two separate times (about a year apart), because *some* of the microarrays came out bad. So instead of running the same samples again (technical replicates), we generated an entirely new set of RNA. (i.e. for treatment A: 1st run- A1, A2, A3; 2nd run- a1, a2, a3) (and for treatment B: 1st run- B1,B2,B3; 2nd run- b1,b2,b3) I was wondering how to handle this sort of data in limma. Because they are biologically distinct, I am tempted to group them together (e.g. combining samples A1,A2,A3 and a1, a2, a3 into A1-A6). But I have reservations about grouping arrays which were run one year apart from each other. How can I prove to myself that samples which were identically treated but came from separate experiments can be grouped together in limma? Do I calculate the correlation coefficient between groups (corfit$consensus) and use that to determine if the groups are closely correlated? or is there another function in limma that can handle this scenario? Any input will be highly appreciated! Edmund Chang

Hi Edmund, Edmund Chang wrote: > Dear list, > I am working with Affymetrix microarrays and looking for differential > gene expression using limma. We repeated the same microarray experiment > two separate times (about a year apart), because *some* of the > microarrays came out bad. So instead of running the same samples again > (technical replicates), we generated an entirely new set of RNA. > (i.e. for treatment A: 1st run- A1, A2, A3; 2nd run- a1, a2, a3) > (and for treatment B: 1st run- B1,B2,B3; 2nd run- b1,b2,b3) > > I was wondering how to handle this sort of data in limma. Because they > are biologically distinct, I am tempted to group them together (e.g. > combining samples A1,A2,A3 and a1, a2, a3 into A1-A6). But I have > reservations about grouping arrays which were run one year apart from > each other. You almost certainly will need to add an effect to your model to account for the two batches of chips. A very good example can be found here: http://bioinf.wehi.edu.au/marray/jsm2005/lab5/lab5.html Note that they are fitting an effect to account for the day that the chips were processed which would be analogous to what you need to do. An alternative would be to fit a mixed model by using the block argument to lmFit(). > > How can I prove to myself that samples which were identically treated > but came from separate experiments can be grouped together in limma? Do > I calculate the correlation coefficient between groups > (corfit$consensus) and use that to determine if the groups are closely > correlated? A relatively easy way to check this would be to do a principal components analysis and plot the first two PCs (you can use plotPCA() in affycoretools to do so). If the separate batches are grouping together, this gives you some assurance that there isn't a batch effect. Personally, I would err on the side of caution and assume that you need to do something to account for differences between the batches. HTH, Jim > > or is there another function in limma that can handle this scenario? > > Any input will be highly appreciated! > Edmund Chang > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor -- James W. MacDonald, M.S. Biostatistician Affymetrix and cDNA Microarray Core University of Michigan Cancer Center 1500 E. Medical Center Drive 7410 CCGC Ann Arbor MI 48109 734-647-5623