Dear limma users and experts, I'm working with two color Agilent GE microarrays with a common reference set-up. Here, I'm using the avereps function to average multiple probes over a specific gene before fitting a linear model to the data. However, I'm getting low FDR corrected p-values over my whole data resulting in almost no differential expression. Trying to circumvent this, I've read that by applying the intraspotCorrelation function and performing a separate channel analysis, you can get better corrected p-values if the corr < 0.5 (as you are now also incorporating the information available in the A-values). I've done this and this is resolving the problem. My question specifically is now if I should do the intraspotCorrelation on the MA object after averaging the probes over a specific gene (on which the linear model will be based), or on the MA object before averaging? I think I should do the intraspotCorrelation function on the same MA object as the linear model will be fitted, but not sure. The two scenarios: 1. Using MA object after averaging over gene ID MA_norm<-normalizeBetweenArrays(MA_norm, method=between) MA_ave<-avereps(MA_norm, ID=MA_norm$genes$ID) corfit_Ave<-intraspotCorrelation(MA_ave, design) fit_ave<-lmscFit(MA_ave, design, correlation=corfit_Ave$consensus) eb_Ave<-eBayes(fit_Ave) 2. Using MA object before averaging over gene ID and using two different MA objects for the correlation estimate and linear model MA_norm<-normalizeBetweenArrays(MA_norm, method=between) MA_ave<-avereps(MA_norm, ID=MA_norm$genes$ID) corfit<-intraspotCorrelation(MA_norm, design) fit<-lmscFit(MA_ave, design, correlation=corfit$consensus) eb<-eBayes(fit) Any advice on this would be very welcome, thanks in advance, Rob. -- output of sessionInfo(): R version 3.0.1 (2013-05-16) Platform: x86_64-w64-mingw32/x64 (64-bit) locale: [1] LC_COLLATE=Dutch_Belgium.1252 LC_CTYPE=Dutch_Belgium.1252 [3] LC_MONETARY=Dutch_Belgium.1252 LC_NUMERIC=C [5] LC_TIME=Dutch_Belgium.1252 attached base packages: [1] stats graphics grDevices utils datasets methods base loaded via a namespace (and not attached): [1] tools_3.0.1 -- Sent via the guest posting facility at bioconductor.org.