I think thats a reasonable approach to use to set up the normalization. The problem here might be that you have too few replicates to be able to fit an effect for both treatment and normalization fraction based on your experimental design. Jeff On Tue, Aug 9, 2016 at 12:09 PM daniel.gaston [bioc] < noreply@bioconductor.org> wrote: > Activity on a post you are following on support.bioconductor.org > > User daniel.gaston <https: support.bioconductor.org="" u="" 10945=""/> wrote Question: > Ballgown Custom Model Question <https: support.bioconductor.org="" p="" 85973=""/>: > > > I'm working on some RNA analysis of a viral system (human infected cells). > I have used HiSat2 and StringTie and am now doing the analysis with > Ballgown. In this set up I have two biological replicates of each condition > and two variables. There is the treated versus control group and then I > have two RNA fractions from each sample (polysome-associated mRNA fraction > and total mRNA fraction). With other methods used my collaborators have > typically used the total mRNA fraction to normalize the polysome-associated > fraction. > > In my case I was looking at setting up this comparison using a linear > model. For right now I did: > > > > mod = model.matrix(~ pData(bg)$fraction + pData(bg)$treatment) > > mod0 = model.matrix(~ pData(bg)$treatment) > > > > Is this reasonable or should I look at implementing a different sort of > normalization method? Perhaps something using the standard libadjust first > on each sample and then within a sample/treatment combo just dividing the > adjusted FPKM values of the polysome fraction by the total? (Which would be > similar to what colleages had done previously, although without the library > adjustment bit) > ------------------------------ > > Post tags: ballgown, rnaseq > > You may reply via email or visit Ballgown Custom Model Question >