Construction of coexpression networks from multiple datasets

Question

[druggable]

Hi Everyone,

I plan to construct coexpression networks to find modules using various independent datasets from databases such as GEO. Would it be valid that for each dataset, I calculate the fold changes of all genes with respect to untreated samples. Then I combine all the multiple datasets to calculate the coexpression. Or would I need to do some form of normalization first to account for different datasets before calculating the fold changes?

Thanks for your help.