Threshold in WGCNA

Question

[Arindam Ghosh]

In WGCNA there are 3 major steps:

1. Correlation calculation between genes
2. Adjacency matrix generation by raising correlation matrix to power β (soft threshold)
3. TOM based dissimilarity matrix generation for identifying modules/clusters

If I understand correctly, β is selected such that the resulting network fits scale-free topology. Till step 2, we will have all-to-all connection between the genes. The adjacency matrix in step 2 gives for the weight for each edge. At this stage how does it select which edges to retain in the final network?

Answer 1

[andres.firrincieli]

The adjacency matrix in step 2 gives for the weight for each edge. At this stage how does it select which edges to retain in the final network?

In WGCNA you do not select which edges must be retained in the final network. However, when you raise the correlation matrix to power β you are suppressing edges with low correlations values: link

edit: There is a way to filter out genes with low connectivity Check the SI of this paper