Hi,
I'm analyzing a dataset with DESeq, which consists of two time-points and two genotypes. I am not very knowledgeable in statistics, sadly, and I am a bit lost as to whether I should use DESeq under the "standard" comparison mode, or make use of the more "advanced" GLM regressions facet.
The comparisons are between :
GenotypeA/Timepoint1 VS GenotypeA/Timepoint2
GenotypeB/Timepoint1 VS GenotypeB/Timepoint1
GenotypeA/Timepoint1 VS GenotypeB/Timepoint1
GenotypeA/Timepoint2 VS GenotypeB/Timepoint2
My understanding is that although my experimental design involves two factors (genotype and timepoint), since my comparisons are only changing one factor at a time, I have nothing to do performing Linear Regressions / Model Fitting, but I am not sure of this. Perhaps it would make sense to use the multi-factorial design analysis ONLY if I were to compare GenotypeA/Timpepoint1 VS GenotypeB/Timepoint 2? (This changing two factors at a time in a single comparison)
Also, what if I want to create a PCA plot of the libraries? Is that a sensible reason to use a GLM fit?
Thanks, and sorry for the extreme confusion.
Carmen
Although they are written for different software packages, you should read the edgeR and limma user's guides. There are many examples that show you how to apply linear models to different experimental designs, some of which are very similar to yours.
Chapter 8, in particular, in the limma user's guide is extremely helpful. Play close attention to Chapter 8.5 (Interaction Models: 2 × 2 Factorial Designs) for a simpler approach to analyzing multi-factorial experiments.
Thanks for the recommendation, again, Steve! I hope I can brush a bit on my understanding on the statistics of this. I'll go read now...
Log in to answer this question.
