Hi all,
I have been asked to give advice on designing an RNAseq experiment with follow-up measurements and since I'm a statistician who recently started working on analysis of data from RNAseq experiments I would greatly appreciate any feedback you may have.
In particular, 200 patients (100 cases and 100 controls) will be followed-up in time and RNAseq data will be available at 5 time points per patient. The sequencing will be done on the platform HiSeq2500 from Illumina. The main questions of interest are: (1) detect differentially expressed genes between cases and controls at specific points in time and (2) identify genes with expression patterns in time different between cases in controls (i.e. time by group interaction term(s)).
My thoughts are as follows:
- To address question (2) my advice is of course to make sure that the whole profile of each patient (i.e. all 5 measurements) are placed on the same 96-well plate. Thus we can study changes in time which will not be confounded by the plate effect.
- To address question (1) we need to have 50% of the profiles from the cases and the other 50% from the controls (50-50% as much as possible) on the same 96-well plate. Thus we can test for differences between cases and controls without any plate confounding.
- I have read several posts on this list about the desired coverage and my conclusion is that approximately 20 million reads per sample is acceptable (single-end is most probably what we will do at the end). This means that provided that a lane can give you at most 200million reads then you may put at most 10 samples per lane. i.e. the 5 measurements of a case and the 5 measurements of a control. The sequencing team told me that they plan to fill in the whole 96-well plate i.e. approximately 18 profiles: 18*5 + 3 = 93 samples on the 96-well plate (the 3 samples will be used as standards) and then they will place the material on many lanes until the desired coverage is reached. I think the 18 profiles are a bit too much and I'm not sure if the desired coverage will be reached. However since I'm not very experienced on this I would greatly appreciate your thoughts on that.
Kind regards,
Jane
What I don't understand above is how will the whole thing will work. What is the role of these 96 well plates long term? Will the plates be shuffled constantly in and out of the freezer as they are filled up? Frankly that sounds like it has the potential for trouble.
You should definitely distribute all samples across all lanes. In general I will also say that the more samples the more things can and will go wrong. Things like mislabeling, adding wrong barcodes etc.