Dear list, This question might be a bit redundant and I apologize for it, if it is, but I can't find a clear answer to what I'm trying to do. I am working on a set of 12 expression one-channel arrays. My target file is basically as follows: FileName Pairs Groups Sample1 1 Group1 Sample2 1 Group1 Sample3 1 Group2 Sample4 1 Group2 Sample5 1 Group3 Sample6 1 Group3 Sample7 2 Group1 Sample8 2 Group2 Sample9 2 Group3 Sample10 3 Group1 Sample11 3 Group2 Sample12 3 Group3 There are several parameters to take into account: - I want to produce all possible pairwise comparisons (Group3-Group2, Group2-Group1, Group3-Group1): "Groups" column - I want my design to take into account the paired samples: "Pairs" column - The last thing is that some samples are technical replicates (Sample1 with Sample2, Sample3 with Sample4, Sample5 with Sample6) and I would also like to take this into account. I've read the "8.7 Multi-level experiments" chapter of limma user guide, which guided me into combining paired data and multiple comparisons, in which case I would do: > design <- model.matrix(~0+factor(targets$Groups)) > colnames(design) <- unique(targets$Groups) > corfit <- duplicateCorrelation(eset,design,block=targets$Pairs) > fit <- lmFit(eset,design,block=targets$Pairs,correlation=corfit$consensus) In theory to take into account technical replicates I would use: > biolrep <- c(1,1,2,2,3,3,4,5,6,7,8,9) > corfit <- duplicateCorrelation(eset, block = biolrep) > fit <- lmFit(eset, block = biolrep, cor = corfit$consensus) But how can I combine all of this? Is there a way to somehow pass both paired and technical replication information into the duplicateCorrelation step? Or should I modify the design instead to take into account the paired design? It is getting quite confusing for me. Any help greatly appreciated! Thanks in advance! [[alternative HTML version deleted]]