search

llrs / blogR

My blogdown site with R related posts
https://llrs.dev
0 stars 0 forks source link

MA plots and design matrix #52

Open llrs opened 3 years ago

llrs commented 3 years ago

Compare corrections using different models, even if the models result in the same contrast (thus to me equal), the result is not the same:

Check this thread with the explanation and pictures.

Might be worth for:

llrs commented 3 years ago

Probably should explain why I consider this important. I think (need to check that) that this explains something important about the data itself. If splitting a variable on multiple factors results in incorrect normalization, then this variable doesn't really have those much factors. Probably this has been said more elegantly and mathematically correctness but I haven't absorbed/found it yet.

Should probably check the maths of voom function to see if it is not really a mathematical construct or consequence.

llrs commented 3 years ago

Further idea for the package: select which samples ones wants on each side and then build the model that allow to make those contrasts.

I want A1, A2, A3 vs B1, B2, B3 for comparison 1 (A vs B) and A1, B1, A3, B3 vs A2, B2 (odds vs even) on comparison 2, and then ok, the model must be that way: ~0+A+B and comparisons defined as: A -B for comparison 1 and ??? for comparison 2. Or just say that this is not possible on a single model of the data.