I think perhaps I was just getting confused. The function modelMatrix() throws an error message unless the number of columns you want in the design matrix is one less than the number of RNA targets. This led me to get confused as to how you could create all contrasts of interest if one was missing from the design matrix. However, if we have four RNA targets (R1-4) in a loop design, we can choose arbitrarily to use one as the ref: modelMatrix(Targets, ref="R1") Then apply our contrasts of interest R2-R3, R3-R4, R2-R4 and then any comparisons to R1 are simply R2, R3 and R4.... Correct? Mick -----Original Message----- From: Gordon Smyth [mailto:smyth@wehi.edu.au] Sent: 10 December 2005 11:27 To: michael watson (IAH-C) Cc: bioconductor at stat.math.ethz.ch Subject: [BioC] Limma and creating design matrices for complex loop designs >Date: Fri, 9 Dec 2005 11:21:49 -0000 >From: "michael watson $IAH-C$" <michael.watson at="" bbsrc.ac.uk=""> >Subject: [BioC] Limma and creating design matrices for complex loop > designs >To: <bioconductor at="" stat.math.ethz.ch=""> > >Hi > >I would like to respectfully request more examples of the creation of >design matrices for complex (ie involving more than 3 samples) loop >designs using two-colour microarrays in the limma documentation. > >Although section 20.5 does give an example, with 5 RNA samples, one of >these is a pool which makes an obvious reference when creating the >design matrix as it is a coefficent we are not interested in estimating. >However, other loop designs may not contain such an obvious sample, and >I am struggling to figure out a way to create a design matrix for my >current design that allows all contrasts of interest. In fact, at >present, I have to create one design matrix to get some of the >contrasts and another to get the rest. Is this a valid approach? Valid if both design matrices span the full treatment space, but never necessary. Gordon >Many thanks > >Mick