Dear all, I have an RNA-seq experiment where I want to test for differential expression in response to my applied treatment. As biological replicates I have two different genotypes of my clonal species, which were each exposed to treated and untreated conditions. The straight forward way to test for a treatment effect would therefore be: genotype <- as.factor(c("g1","g1","g2","g2")) treat <- as.factor(c("U","T","U","T")) design <- model.matrix(~treat) design (Intercept) treatU 1 1 1 2 1 0 3 1 1 4 1 0 ... lrt.tagd_treat <- glmLRT(D, glmfit.tagd, coef=2) topTags(lrt.tagd_treat) However, when I am looking at the MDS plot of my 4 samples I can see the the effect of the genotype is also not neglectable. Would it therefore make sense to include the factor genotype in the design matrix as well to adjust for the genotype effect in my model, e.g.: design <- model.matrix(~treat+genotype) design (Intercept) treatU genotypeg2 1 1 1 0 2 1 0 0 3 1 1 1 4 1 0 1 ... lrt.tagd_treat <- glmLRT(D, glmfit.tagd, coef=2) topTags(lrt.tagd_treat) 1) Is an adjustment for an "unwanted effect" possible this way in general? 2) Does it also make sense in my case regarding the very low level of biol. replication (-> 2)? Thanks a lot, Lina [[alternative HTML version deleted]]

Dear Lina, Responses interpolated below. > Date: Fri, 17 Feb 2012 12:56:13 +0100 > From: Lina Weber <linamweb at="" googlemail.com=""> > To: bioconductor at r-project.org > Subject: [BioC] edgeR: GLM - adjusting for unwanted effect?! > > Dear all, > > I have an RNA-seq experiment where I want to test for differential > expression in response to my applied treatment. As biological replicates I > have two different genotypes of my clonal species, which were each exposed > to treated and untreated conditions. > > The straight forward way to test for a treatment effect would therefore be: > > genotype <- as.factor(c("g1","g1","g2","g2")) > treat <- as.factor(c("U","T","U","T")) > > design <- model.matrix(~treat) > design > (Intercept) treatU > 1 1 1 > 2 1 0 > 3 1 1 > 4 1 0 > > ... > > lrt.tagd_treat <- glmLRT(D, glmfit.tagd, coef=2) > topTags(lrt.tagd_treat) > > However, when I am looking at the MDS plot of my 4 samples I can see the > the effect of the genotype is also not neglectable. Would it therefore make > sense to include the factor genotype in the design matrix as well to adjust > for the genotype effect in my model, e.g.: > > design <- model.matrix(~treat+genotype) > design > (Intercept) treatU genotypeg2 > 1 1 1 0 > 2 1 0 0 > 3 1 1 1 > 4 1 0 1 > > ... > > lrt.tagd_treat <- glmLRT(D, glmfit.tagd, coef=2) > topTags(lrt.tagd_treat) > > 1) Is an adjustment for an "unwanted effect" possible this way in general? Yes, this a standard technique. > 2) Does it also make sense in my case regarding the very low level of biol. > replication (-> 2)? You have one degree of freedom left over for estimating the dispersion. It's the minimum, and it would be better to have more, but it is usable. Best wishes Gordon > Thanks a lot, > Lina > ______________________________________________________________________ The information in this email is confidential and intend...{{dropped:4}}