Hello, Sorry that I bring up this again, but I do want to know if my logic is correct or not, with regard to subset a big data set according to different research questions and start from normalization to set up different design matrixes and contrasts. I am particularly unsure about the 2nd question of comparing normal and tumor type I (AR+), and how pairing and batch effect should be addressed? Any comments or suggestions would be very appreciated. Thank you very much in advance. Thanks, Xiayu -----Original Message----- From: bioconductor-bounces@r-project.org [mailto:bioconductor- bounces@r-project.org] On Behalf Of Rao,Xiayu Sent: Monday, July 28, 2014 4:55 PM To: bioconductor at r-project.org Subject: [BioC] multi-level design and contrasts problem - limma Hello, I read the limma user guide on the topics of multi-level experiments and found the information very useful. But my design is a little more complicated, and I would like to consult for a solution. I was asked to solve the following questions regarding the data structure below (targets.txt). I guess I need to set up different design matrixes according to different questions? 1) Normal vs tumor: Do I subset the data into paired samples (subject) only and then used the paired design since some samples do not have their normal samples? There is only 1 or 2 patients with Tumor and normal samples in different chips. Can I just do pairing and ignore the batch effect (chip), as I read in the forum that doing both does no good since most pairs are within the same chip. 2) Normal vs AR positive tumor: Only tumor samples have AR information. I am thinking to pool type and AR together into 1 column called type_AR with 3 categories: tumorNeg, tumorPos, and normal. I will use design <- model.matrix(~subject+type_AR) and set contrasts normal-tumorPos for (2) and normal-tumorNeg for (3). Or I should follow the multi-level design instructions to include the type_AR and chip in the design (paste the two), and then use duplicateCorrelation() on subject? I will ignore gender. 3) Normal vs AR negative tumor: same above. 4) AR positive vs AR negative tumor: I am thinking to remove all normal samples and ignore type, subject and gender. The design would be = model.matrix(~chip+AR), right? 5) Male AR positive vs Female AR positive: One way is to remove all normal and AR negative samples (only gender and chip left), and compare Female and Male using design <- model.matrix(~chip+gender). The 2nd way is to follow multi-level design instructions to allow more comparisons (including AR negative): Treat <- factor(paste(targets$gender,targets$AR,sep=".")) design <- model.matrix(~0+Treat) duplicateCorrelation(eset,design,block=targets$chip) Please let me know if I am on the right track. Thank you very much! Targets.txt: sample type subject gender AR chip s1 tumor 1 M neg 1 s2 normal 1 M 1 s3 tumor 2 M pos 1 s4 normal 2 M 1 s5 tumor 3 F neg 1 s6 normal 3 F 1 s7 tumor 4 M pos 1 s8 normal 4 M 1 s9 tumor 5 M pos 2 s10 normal 5 M 2 s11 normal 6 F 2 s12 tumor 7 M pos 2 s13 normal 7 M 2 s14 tumor 8 M pos 2 s15 normal 8 M 2 s16 tumor 9 M neg 3 s17 tumor 10 M neg 3 s18 tumor 11 F neg 3 s19 tumor 6 F pos 3 s20 tumor 12 F pos 3 s21 tumor 13 F neg 3 s22 tumor 14 F pos 3 Thanks, Xiayu [[alternative HTML version deleted]] _______________________________________________ Bioconductor mailing list Bioconductor at r-project.org https://stat.ethz.ch/mailman/listinfo/bioconductor Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor