Hi, I am analysing single-channel miRNA microarray data: I have 4 treatments at 4 time points, mostly in triplicate, spread approximately evenly across 2 batches. Previously I have used ComBat to reduce batch effects and then used Limma and AgiMicroRna to get DE miRNAs. However, I read on the list somewhere that it is better to add Batch as a factor to the model in Limma rather than try and remove it. ### So this time I added Batch to my targets file...### FileName Treatment GErep Subject Time Compound Batch A1.txt A 1 1 hr4 v 1 A2.txt A 1 2 hr4 v 2 A3.txt A 1 3 hr4 v 2 B1.txt B 2 1 hr4 m 2 B2.txt B 2 2 hr4 m 1 B3.txt B 2 3 hr4 m 1 ... (I also have hr8, hr24, hr120 and Compound b and d) ###And did the following...### TS<-paste(phenoData$Time,phenoData$Compound,phenoData$Batch,sep=".") TS<-factor(TS,levels=c("hr4.v.1","hr4.m.1","hr4.b.1","hr4.d.1", hr4.v.2","hr4.m.2","hr4.b.2","hr4.d.2", "hr8.v.1","hr8.m.1","hr8.b.1","hr8.d.1", "hr8.v.2","hr8.m.2","hr8.b.2","hr8.d.2", "hr24.v.1","hr24.m.1","hr24.b.1","hr24.d.1", "hr24.v.2","hr24.m.2","hr24.b.2","hr24.d.2", "hr120.v.1","hr120.m.1","hr120.b.1","hr120.d.1", "hr120.v.2","hr120.m.2","hr120.b.2","hr120.d.2")) design<-model.matrix(~0+TS) fit<-lmFit(minimalSet,design) ### I now get an error message saying that coefficients are not estimable (partial NA coefficients for 260 probes). Last year I ran the same process but without batch and it worked fine. I would like to find miRNAs that have changed compared to the control at each time point, having taken into account the effect of batch. Any suggestions as to how I can get it to work? Thanks very much Corrinne (Additionally, I am doing the analysis in R development version as I would like to be able to get CI for log fc) [[alternative HTML version deleted]]