filtering genes before or after the correlation?

Question

[tyasird]

I have a gene expression data with 7k genes, and I benchmark this data using gold standards like CORUM (2k genes). It is important because I draw precision-recall curve.

what I do is, I apply correlation to 7k (7k x 7k matrix) then filter genes based on the CORUM (2k x 2k) and then create a network.

I was wondering is this good way to do it, or I just need to filter genes before and continue to analysis.