Thanks Gordon When I called duplicateCorrelation, I got three warning messages: Too much damping - convergence tolerance not achievable in: glmgam.fit( etc etc) My corfit$consensus was then positive, even though it is a dye-swap experiment. Presumably this means I don't have to treat the technical replicates in any special way...? Mick -----Original Message----- From: Gordon Smyth [mailto:smyth@wehi.edu.au] Sent: 11 September 2005 08:31 To: michael watson (IAH-C) Cc: bioconductor at stat.math.ethz.ch Subject: [BioC] Dealing with technical replication with a common reference design >Date: Fri, 9 Sep 2005 09:09:02 +0100 >From: "michael watson \(IAH-C\)" <michael.watson at="" bbsrc.ac.uk=""> >Subject: [BioC] Dealing with technical replication with a common > reference design >To: <bioconductor at="" stat.math.ethz.ch=""> > >Hi > >I have to say the limma userguide is now one of the most comprehensive >guides available for microarray analysis. It is therefore regretfully >that I ask for additional information... > >The guide rightly points out that we must not treat arrays that are >technical replicates as true replicate arrays as they are not >independent. Sections 10.1 and 10.2 then go on to explain how >technical replicates can be handled using either duplicateCorrelation >or by manipulating the contrasts matrix. However, both these examples >deal with arrays where direct comparisons have been made. > >I have data from a common reference design, comparing two groups using >a common reference, with biological replicates and technical replicates >(dye swaps) and it's not immediately clear from the examples how I >should deal with the technical replication. The treatment of technical replicates is the same for common reference designs as it is for direct designs. If all your technical reps are in dye-swap pairs then the example in Section 10.1 with dye-swap technical replicates will carry over to your situation with very few changes. Gordon >Many thanks for any hints forthcoming! > >Mick