Hi I have a huge dataset from TCGA ( many disease cohort) but I'm only interested to compare a couple of diseases. My question is, during normalization should I just normalize the groups I plan to compare or is it better to normalize all the samples ( TMM ) follow with voom with a null design?

thanks!

If the dataset is huge, then you may as well just TMM the samples you plan to use, but it won't make much difference either way.

I'm not sure what you mean by voom with a null design. voom should always be run with the appropriate design matrix. Again, you only need to run voom on the samples you plan to compare.

You might not even need voom. Would limma-trend be just as appropriate? See the comments in the limma User's Guide.