I recently got a new project to work on regarding differential expression analysis of RNA-seq data, and it's slightly different to what I've done previously in that it has three conditions, rather than just two. It's basically three isogenic cell lines (with several replicates each) that differ only by their expression levels of a single, important gene (which has been verified experimentally by other collaborators). The first cell line (A) has very low expression of this gene, the second one (B) is moderately expressed and the third one (C) is highly expressed (relatively speaking). The question is which genes (if any) are differentially expressed in regards to this more-or-less linear single-gene expression variation.
I have so far only worked on simple two-condition DE-analysis (what the edgeR documentation terms "classical" analysis), and so I haven't really experienced this case before. I have not used GLMs before, and I'm wondering what the appropriate experimental design for this sort of question would be. I could do the "classical" analysis for A vs. B and A vs. C, for example, and see if the same DEGs from A-B are more differentially expressed in the A-C case. I have done this in a preliminary way, and it does indeed seem to be the case that most of the DEGs follow the trend of higher expression --> higher fold change. It occured to me, however, that this GLM-thing I heard about might be better suited. Seeing as I haven't used it before, I quickly skimmed the edgeR document, concluded that I indeed lacked knowledge of the general theory, and went here.
What is the experimental design I'm looking for here? Should I read up more carefully on GLMs and their use, or do I stick with the classical analysis? If GLMs are the way to go, is there a "beginner's guide"-kind of explanation somewhere for somebody new to the idea? While I've certainly learnt a lot of statistics during my PhD, yet I have little formal training, so I'd like to get an intuitive concept down before I dig into the details.
I have so far only worked on simple two-condition DE-analysis (what the edgeR documentation terms "classical" analysis), and so I haven't really experienced this case before. I have not used GLMs before, and I'm wondering what the appropriate experimental design for this sort of question would be. I could do the "classical" analysis for A vs. B and A vs. C, for example, and see if the same DEGs from A-B are more differentially expressed in the A-C case. I have done this in a preliminary way, and it does indeed seem to be the case that most of the DEGs follow the trend of higher expression --> higher fold change. It occured to me, however, that this GLM-thing I heard about might be better suited. Seeing as I haven't used it before, I quickly skimmed the edgeR document, concluded that I indeed lacked knowledge of the general theory, and went here.
What is the experimental design I'm looking for here? Should I read up more carefully on GLMs and their use, or do I stick with the classical analysis? If GLMs are the way to go, is there a "beginner's guide"-kind of explanation somewhere for somebody new to the idea? While I've certainly learnt a lot of statistics during my PhD, yet I have little formal training, so I'd like to get an intuitive concept down before I dig into the details.
