To Whom it may Concern, I'm would like to inquire about the appropriate model design in limma for the model below. All patients are healthy, female and lean. Patient Tissue 1 Blood 1 Adipose Tissue 1 Buccal Cells 1 Lymph node 2 Blood 2 Adipose Tissue 2 Buccal Cells 2 Lymph Node 3 Blood 3 Adipose Tissue 3 Buccal Cells 3 Lymph Node We have implemented 2 different ways to implement this: first is simple "pairing" of samples: design <- model.matrix(~Patient + Tissue) fit <- lmfit(data, design) The second treats the data as biological replicates, and uses "Patient" as a blocking variable instead: design <- model.matrix(~0+Tissue) corfit <- duplicateCorrelation(data, design, block=Patient) fit <- lmFit(data, design, block=Patient, correlation=corfit$consensus) The two different methods give different results. We have read section 8.7 in the limma users guide and are still unsure which method is more appropriate for this particular experimental design. Would you be able to give feedback on this please? Regards, Dr. Tim Peters Postdoctoral Fellow - Bioinformatics and Statistics CSIRO Mathematics, Informatics and Statistics Ph: +61 2 9325 3266 | Email: tim.peters@csiro.au Address: Riverside Life Sciences Centre, 11 Julius Avenue, North Ryde, NSW 2113, Australia [[alternative HTML version deleted]]

Hi Tim, On 1/29/2013 8:02 PM, Tim.Peters at csiro.au wrote: > To Whom it may Concern, > > I'm would like to inquire about the appropriate model design in limma for the model below. All patients are healthy, female and lean. > > Patient > > Tissue > > 1 > > Blood > > 1 > > Adipose Tissue > > 1 > > Buccal Cells > > 1 > > Lymph node > > 2 > > Blood > > 2 > > Adipose Tissue > > 2 > > Buccal Cells > > 2 > > Lymph Node > > 3 > > Blood > > 3 > > Adipose Tissue > > 3 > > Buccal Cells > > 3 > > Lymph Node > > > We have implemented 2 different ways to implement this: first is simple "pairing" of samples: > > design<- model.matrix(~Patient + Tissue) > fit<- lmfit(data, design) > > The second treats the data as biological replicates, and uses "Patient" as a blocking variable instead: > > design<- model.matrix(~0+Tissue) > corfit<- duplicateCorrelation(data, design, block=Patient) > fit<- lmFit(data, design, block=Patient, correlation=corfit$consensus) > > The two different methods give different results. We have read section 8.7 in the limma users guide and are still unsure which method is more appropriate for this particular experimental design. Would you be able to give feedback on this please? I don't think section 8.7 is necessarily applicable here. Your table got destroyed in transmission, but it appears you have all tissues measured on all patients, so there is no nesting here. I would argue that parsimony and simplicity are guiding principles, and would thus favor the simple paired t-test approach. Fewer and simpler assumptions are usually better. Best, Jim > > Regards, > > Dr. Tim Peters > Postdoctoral Fellow - Bioinformatics and Statistics > CSIRO Mathematics, Informatics and Statistics > Ph: +61 2 9325 3266 | Email: tim.peters at csiro.au > Address: Riverside Life Sciences Centre, 11 Julius Avenue, North Ryde, NSW 2113, Australia > > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor -- James W. MacDonald, M.S. Biostatistician University of Washington Environmental and Occupational Health Sciences 4225 Roosevelt Way NE, # 100 Seattle WA 98105-6099