I have a question about Paired analysis in EdgeR. I have read the Users Guide (Aug 2013) and this clearly describes several types of paired analysis and how to build the appropriate design matrixes. However, the design closest to my experiment (4.2, p40) doesn't seem to be a paired analysis. My experiment is as follows: "cases" - placentas from patients with a well defined poor obstetric outcome, "controls" placentas from 20 subjects with good obstetric outcome. The controls were selected from a large cohort (~4000) to be a close match to the cases. Matching was based on ~20 characteristics (maternal age, BMI, gestational age, fetal sex, mode of delivery etc etc). Hence there is 1 to 1 matching of cases and controls and hence a paired design. We collected RNA-Seq data from all 40 - individually bar-codded, pooled and run on 3 lanes of a HiSeq2000. I am probably being a bit slow, but advice on what the design matrix should look like would be a real help. Thanks. Steve -- output of sessionInfo(): -- Sent via the guest posting facility at bioconductor.org.

Hi Steve, The simple answer is that you don't have a paired design. Simply matching cases and controls is not the same thing. The reason people account for pairing (or other dependence structures) is because paired data violate the assumption of independence between samples that one normally makes when fitting a linear model. In other words, we normally assume that there is no correlation between samples, and have to make adjustments when there is some correlation. Matching cases and controls on a set of phenotypes doesn't introduce correlation. Instead, it is simply a method that tries to reduce unwanted variability from uninteresting phenotypes. Best, Jim On Thursday, October 31, 2013 9:43:53 AM, Steve [guest] wrote: > > I have a question about Paired analysis in EdgeR. I have read the Users Guide (Aug 2013) and this clearly describes several types of paired analysis and how to build the appropriate design matrixes. However, the design closest to my experiment (4.2, p40) doesn't seem to be a paired analysis. > > My experiment is as follows: "cases" - placentas from patients with a well defined poor obstetric outcome, "controls" placentas from 20 subjects with good obstetric outcome. The controls were selected from a large cohort (~4000) to be a close match to the cases. Matching was based on ~20 characteristics (maternal age, BMI, gestational age, fetal sex, mode of delivery etc etc). Hence there is 1 to 1 matching of cases and controls and hence a paired design. We collected RNA-Seq data from all 40 - individually bar-codded, pooled and run on 3 lanes of a HiSeq2000. > > I am probably being a bit slow, but advice on what the design matrix should look like would be a real help. > > Thanks. > > Steve > > > -- output of sessionInfo(): > > > > -- > Sent via the guest posting facility at bioconductor.org. > > _______________________________________________ > 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 -- James W. MacDonald, M.S. Biostatistician University of Washington Environmental and Occupational Health Sciences 4225 Roosevelt Way NE, # 100 Seattle WA 98105-6099