This has been done previously in 2 studies (possibly many more):

• Secreted frizzled-related protein 1 loss contributes to tumor phenotype of clear cell renal cell carcinoma
• Pathway signature and cellular differentiation in clear cell renal cell carcinoma

Both of these utilised this GEO dataset: GSE6344

The methods that they used for merging, which involves the determination of a scaling factor on raw signal data, are here: ftp://ftp.ncbi.nlm.nih.gov/geo/series/GSE6nnn/GSE6344/suppl/GSE6344_MethodsDetails.PDF

This topic has also been discussed on both Biostars ( Combining Two Platforms Affy Hgu133A And Hgu133B ) and Bioconductor a few times.

[just briefing everyone]

--------------------------------

The key to your particular question is that you just want to run WGCNA on the data. As WGCNA is fundamentally based on correlation, in this case, you just need to process each array experiment independently and then combine the final normalised datasets together. The merged dataset then becomes the input to WGCNA. Obviously only combine the datasets where there are common genes.

An optional transformation for you is to normalise the arrays independently, covert the normalised expression values to the Z-scale, and then merge everything together.

If you want to start conducting statistical analyses other than WGCNA, then it's a different story and that particular topic is hot and provides for debate across the forums.

Kevin