Signed vs unsigned network construction: wgcna issue!

Question

[emanuela.molinari]

Dear all:

I am trying to run WGCNA on my RNA-seq dataset (20 samples, 2 conditions 10 obs each). I had a batch effect corrected with edgeR.

When I try to apply the soft threshold on my dataset I run into the following dilemma: if I set the network type "unsigned" I obtain a soft threshold of 15. If I rather make it "signed" then even after a power of 30 it doesn't get a scale free topology index of 0.9.

My knowledge in statistics is limited, and I cannot understand the reasons for such a weird behavior.

Can anyone help?

Thanks!

Manu

Answer 1

[Brian Gudenas]

A scale free topology index of 0.9 or greater is not required for WGCNA, although it should probably be above 0.8

Just take a look at the R code from this publication (authored by the creator of WGCNA), they use a soft threshold of 26 (signed network) which only achieves an index of 0.808.

http://labs.genetics.ucla.edu/horvath/htdocs/CoexpressionNetwork/developingcortex/

It sounds like the issue is in the connectivity of your RNA-seq data though, is it being filtering by some criteria such as differential expression?