Dear Gordon: Thank you very much for your reply. I think we are going to use moderated t-statistic as you suggested. And we are using external controls and spike to do the normalization. I still wondering whether I could use all the replicates on one slide. 6 of them are regular spacing and spacing is 800. the result comes out pretty reasonable. The commands look like following: design <- c(1,1,1) fit <- lmFit(MA, design, ndups=6, spacing=800) eb <- ebayes(fit) y <- toptable(number=length(fit$coefficients), genelist=RG$genes, fit=fit, A = fit$Amean, eb=eb, adjust="fdr") write.table(y, file="diff_result.txt", sep="\t ") But when I try to use 12 replicates on one slide, the result doesn't look right. What I am trying to do is: (1) export Block, Row, Column, Name, M value and weights to a text file. (2)sort the text file by Gene Name in Excel, so the same genes will appear in adjacent rows (3)using limma to do linear fit, put dups as 12, spacing as1. The commands look like following: result <- data.frame(cbind(MA$M[,1], MA$weights[,1], MA$M[,2], MA$weights[,2], MA$M[,3], MA$weights[,3])) names(result) <- c("Rep1", "Flag1", "Rep2", "Flag2", "Rep3", "Flag3") result <- cbind(MA$genes, result) write.table(result, file="result.txt", sep="\t", row.names=TRUE) After sorting in Excel by Name, x <- read.table("result.txt", sep="\t", as.is=TRUE, header=TRUE, comment.char="") MA <- list(M=cbind(x$Rep1,x$Rep2,x$Rep3), genes=cbind(x$Block,x$Row,x$Column, x$Name), weights=cbind(x$Flag1,x$Flag2,x$Flag3)) MA <- new("MAList", MA) design <- c(1,1,1) fit <- lmFit(MA, design, ndups=12, spacing=1) eb <- ebayes(fit) y <- toptable(number=length(fit$coefficients), genelist=MA$genes, fit=fit, A = fit$Amean, sort.by="T", eb=eb, adjust="fdr") write.table(y, file="rep_result.txt", sep=" ") Is there something wrong when I try to account for 12 replicates in one slide? I highly appreciate your help. Hua ----- Original Message ----- From: "Gordon Smyth" <smyth@wehi.edu.au> To: "Hua Weng" <hweng@bmb-fs1.biochem.okstate.edu> Cc: <bioconductor@stat.math.ethz.ch>; <margess@bmb- fs1.biochem.okstate.edu=""> Sent: Monday, July 19, 2004 7:21 PM Subject: Re: [BioC] Bayesian method in limma > At 01:42 AM 20/07/2004, Hua Weng wrote: > >Dear Gordon: > > > >I tried to use ebayes() method to pick differential expressed genes for our > >two color cDNA microarray data analysis. Our expreiment is designed to > >anaylysis the influense of one chemical on cotton genes. It is a > >Single-Sample Expreiment and two RNA sources are comapred deirectly on 3 > >biological replicates through time course. But we don't have complete cotton > >gene library. So we hand pick around 200 genes that we expect to be > >differentiall expressed and print them 12 times (technical replicates) on > >the same slide. I have following questions: > > > >(1) Based our expreiment design, is it still suitable to use ebayes() method > >to pick differentially expressed genes for our data? > > > >(2)I took a look at help file for ebayes() and found out the default vale > >for proportion is 0.01. What value should I set for proportion for our case? > >Do I need to change default value for stdev.coef.lim? > > You can still use ebayes(), but use the moderated t-statistic rather than > the log-odds (B-statistic) because the former doesn't require an estimate > of the proportion of differentially expressed genes. > > A more subtle question is to normalize data where most or all of the genes > are differentially expressed. > > Gordon > > >I highly appreciate your help. > > > >Hua > > > >Microarray Core Facility > >Department of Biochemistry and Molecular Biology > > > >Oklahoma State University > > > >348E Noble Research Center > > > >Stillwater, OK 74078 > > > > Phone: 405-744-6209 > > > >Fax: 405-744-7799 >