WGCNA Network Analysis

Question

[skjobs]

I'm trying to use WGCNA for weighted network analysis. I'm getting trouble how to decide this parameter. It's clear how to define and why this has been defined.

#Choosing a soft-threshold to fit a scale-free topology to the network

powers = c(c(1:10), seq(from = 12, to=20, by=2));

What is the meaning of this powers and how to decide this value?

[ATpoint]

I suggest to first get an overview of the method, for example with this lecture from the WGCNA developer, which also covers the power parameter.

Answer 1

[Jean-Karim Heriche]

This parameter must be >1 and is simply the power to which the correlations are raised. This increases the contrast between high and low values, in effect a form of soft thresholding. In the paper, the authors suggest setting it so that the resulting network has a scale-free topology. See the pickSoftThreshold() function in the WGCNA R package.

Answer 2

[Renesh]

In WGCNA, power parameter is needed to reduce the spurious correlations in the data. To select a power from pickSoftThreshold function, choose reasonably high R^2 (column 2), usually higher than say .85 and negative slope around -1 (column 4) to get an approximately scale-free network. Power is helpful to easily differentiate strong and weak correlations.