Normalization of methylation EPIC array data before or after subsetting?

Question

[fdav]

Dear all,

I am working on a large DNA methylation dataset (Illumina EPIC array) that comprises 1300 samples from about 900 individuals, with longitudinal measures for 400 individuals. For a differential methylation analysis in a specific subset of this sample (n=600, cross-sectional) I am now wondering which workflow would make more sense:

a) Perform preprocessing and quantile normalization on full dataset, subset afterwards
b) Subset raw data first, perform preprocessing and quantile normalization on subset

For the normalization, minfi::preprocessQuantile() is used.

Is there a recommended way to do this? Thanks for any help!

[Basti]

In my opinion, it is best practice to normalize all the samples you have together and in differential analysis to use contrasts.