Hi all, Unfortunately I cannot find out how limma calculates t-tests for my factorial design of micrarrays, but let's begin from the start: I have the following microarray experiment: WT Knock-Out Treated 2 2 Untreated 2 2 This way I can make the following comparisons: WT Treated / WT Untreated: This shows me, what the effect of the treatment itself is. Knock-out Treated / Knock-out Untreated: To see what effect the treatment has on Knock out animals. (Knock-out Treated / Knock-out Untreated) / (WT Treated / WT Untreated): The effect of treatment to Knock-Out animals under consideration of changes of gene expression due to treatment in WT animals. So my main interest lies in the last comparison. I did this with the following code: targets <-read.table(file="Targets.csv", sep="\t", header=T, row.names=1, as.is=T) TS <- paste(targets$Strain, targets$Treatment, sep=".") TS <- factor(TS, levels=c("wt.non","ko.non","wt.treat","ko.treat")) design <- model.matrix(~0+TS) colnames(design) <- levels(TS) fit <- lmFit(eset, design) cont.matrix <- makeContrasts( WT.nonVStreat =wt.non-wt.treat, KO.nonVStreat=ko.non-ko.treat, Diff=(ko.non-ko.treat)-(wt.non-wt.treat), levels=design) fit2 <- contrasts.fit(fit, cont.matrix) fit2 <- eBayes(fit2) eset contains my expression values in log2 scale. The contrast are calculated by wt.non-wt.treat ko.non-ko.treat (ko.non-ko.treat)-(wt.non-wt.treat) as I described above. For all 3 comparisons, eBayes calculates t-tests, but for the last comparison, which values are used for the t-test? All signals of ko.non and ko.treat compared to wt.non and wt.treat? I think it's calculated in a completely different way, does anyone know how? Best regards, Pascal

Hi Pascal, On 9/10/2010 10:04 AM, Pascal Gellert wrote: > Hi all, > > Unfortunately I cannot find out how limma calculates t-tests for my > factorial design of micrarrays, but let's begin from the start: > > I have the following microarray experiment: > > WT Knock-Out > Treated 2 2 > Untreated 2 2 > > This way I can make the following comparisons: > > WT Treated / WT Untreated: This shows me, what the effect of the > treatment itself is. > Knock-out Treated / Knock-out Untreated: To see what effect the > treatment has on Knock out animals. > (Knock-out Treated / Knock-out Untreated) / (WT Treated / WT Untreated): > The effect of treatment to Knock-Out animals under consideration of > changes of gene expression due to treatment in WT animals. > > So my main interest lies in the last comparison. I did this with the > following code: > > targets <-read.table(file="Targets.csv", sep="\t", header=T, > row.names=1, as.is=T) > TS <- paste(targets$Strain, targets$Treatment, sep=".") > TS <- factor(TS, levels=c("wt.non","ko.non","wt.treat","ko.treat")) > > design <- model.matrix(~0+TS) > colnames(design) <- levels(TS) > fit <- lmFit(eset, design) > > cont.matrix <- makeContrasts( > WT.nonVStreat =wt.non-wt.treat, > KO.nonVStreat=ko.non-ko.treat, > Diff=(ko.non-ko.treat)-(wt.non-wt.treat), > levels=design) > fit2 <- contrasts.fit(fit, cont.matrix) > fit2 <- eBayes(fit2) > > eset contains my expression values in log2 scale. > The contrast are calculated by > wt.non-wt.treat > ko.non-ko.treat > (ko.non-ko.treat)-(wt.non-wt.treat) > as I described above. For all 3 comparisons, eBayes calculates t-tests, > but for the last comparison, which values are used for the t-test? > All signals of ko.non and ko.treat compared to wt.non and wt.treat? I > think it's calculated in a completely different way, does anyone know how? It's not calculated in a completely different way. The contrasts you are computing are just the numerator for the t-statistic. So for the first two contrasts, the numerator is simply the difference between the means of two groups. In the third case, the numerator is the difference between the differences (e.g., is calculated exactly as you have specified). Best, Jim > > Best regards, > > Pascal > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor -- James W. MacDonald, M.S. Biostatistician Douglas Lab University of Michigan Department of Human Genetics 5912 Buhl 1241 E. Catherine St. Ann Arbor MI 48109-5618 734-615-7826 ********************************************************** Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues