When is it valid to use differentially expressed genes as input in WGCNA?

Question

[bren12345]

I have read that is not recommend to use differentially expressed genes as input in WGCNA. However, I have seen some papers where they have use applied WGCNA on DEG.

My question is when is it valid to do this? Can I do this if I did not choose the soft thresholding power by scale-free topology?

[Andrew]

I'm a newbie commenting five years later. Guys with the same confusion can refer to this paper. "Improved biomarker discovery through a plot twist in transcriptomic data analysis" https://doi.org/10.1186/s12915-022-01398-w

Answer 1

[Kevin Blighe]

You can use as input to WGCNA what you please; however, the interpretation that you then make on the data will change.

Typically, one uses the entire dataset because the logic is that WGCNA identifies modules that you then associate with your outcomes and traits of interest. In a sense, WGCNA 'decomposes' your genes into modules, with the statistical inferences then being made on the derived modules. When you just use DEGs as input, you disrupt this logic, and, in this case, WGCNA may not be quite relevant.

Kevin

[bren12345]

Thanks for your reply. I want to use WGCNA to identify the modules with the highest correlation with certain trait, select the genes that belong to such modules and then, perform further analysis. So, is just a way to prioritize some genes from the complete list of differentially expressed genes. Would you consider this an appropriate approach?

[Kevin Blighe]

That approach sounds fine but, if you just use the differentially expressed genes (DEGs) as input to WGCNA, then it is a 'supervised' / 'biased' approach, but possibly no different than doing hierarchical clustering on DEGs in a supervised way.