Hi Consider the design matrix in the limma user guide, section 10.7: FileName Strain Treatment File1 WT U File2 WT S File3 Mu U File4 Mu S File5 Mu S Does it matter if these samples are from five different mice, or only two (WT and Mu) where different treatments have been applied to the same mouse at different times? Does this nesting change how we apply the design and contrasts matrices in limma? Consider also another level, a simple time of "before" and after "treatment". We would have something like: FileName Strain Treatment Time File1 WT U Before File2 WT U After File3 WT S Before File4 WT S After Etc Etc Presumably then the contract "After - Before" would show genes which differ due to the administration of a treatment (regardless of what the treatment is), the contrast "S - U" would show genes changing due to the treatment, and "WT - Mu" would look for strain effects. But "Before" and "After" are relative to the treatment and the mouse, rather than specific (equivalent) moments in time; again, does this change how we apply limma? Thanks in advance for your help Mick

>Date: Mon, 6 Dec 2004 13:41:35 -0000 >From: "michael watson \(IAH-C\)" <michael.watson@bbsrc.ac.uk> >Subject: [BioC] Handling "nested" factors in limma >To: <bioconductor@stat.math.ethz.ch> > >Hi > >Consider the design matrix in the limma user guide, section 10.7: > >FileName Strain Treatment >File1 WT U >File2 WT S >File3 Mu U >File4 Mu S >File5 Mu S > >Does it matter if these samples are from five different mice, or only >two (WT and Mu) where different treatments have been applied to the same >mouse at different times? Yes, it makes a fundamental difference to the conclusions you can make from the experiment. The difference is to do with science rather than with limma or with statistical computations. It means that any conclusions you make from the experiment would apply only to these two mice, not to wildtype and mutant mice in general, because you have not estimated the variability of mice within the two populations. Observing both U and S on each mouse would be a good idea if you had more than one mouse per strain. However the model would then be more complicated -- you would have 3-factors, or rather 2-factors with a blocking structure. >Does this nesting change how we apply the >design and contrasts matrices in limma? No. Gordon