Consensus WGCNA: soft-threshold power selection

Question

[baldauf]

Hi, I just started to get familiar with WGCNA and I am trying to apply the tutorials written by Peter Langfelder and Steve Horvath for the Consensus WGCNA on my data. Can any one tell me how I obtain the powerEstimate of the "pickSoftThreshold" for both of my data sets included in the multiExpr Set, which I build at the beginning according to the tutorial?

I also analyzed both data sets separately to perform step 3 "Relate consensus modules to set-specific modules) of the tutorial (https://labs.genetics.ucla.edu/horvath/CoexpressionNetwork/Rpackages/WGCNA/Tutorials/Consensus-NetworkConstruction-blockwise.pdf). Should I choose the same power in each case, e.g. power 6 for the set-specific WGCNA and power 6 for the consensus module?

Thanks, Jutta

Answer 1

[Kevin Blighe]

To put it bluntly, the soft thresholding power is virtually always 6, and that's coming from the mouth of a person I know who has been teaching WGCNA for years and who worked with Horvath in the past.

When you generate the graphs that plot the powers, you are essentially looking for the inflexion points. In all graphs in the tutorial, it's around the 6, 7, 8 mark. You then choose the lowest number at that [inflexion] point.

That's it explained in layman's terms.

Kevin

Edit February 14th, 2019: generally choose the first power that passes 0.9 on the y-axis.

[baldauf]

Thanks for your response! Does your "virtually always 6" only refers to the data of the tutorials, or do you mean (or the person you know) that for all data sets in general the soft thresholding power is always 6? For my data (see figure: https://ibb.co/eyMHJT ), these inflexion points are not as clearly defined (especially the red one). In the blue one, I would think it is around 7,8,9. So I would choose power =7 ?

How does the selection of the power influence the results later on?

[Kevin Blighe]

For most datasets, it is usually 6. For your data, it looks more like 7 or 8 for both parents and hybrids. I don't believe it influences the overall result, but you should nevertheless repeat the analysis at 6, 7, and 8 in order to check.

[baldauf]

Ok, thank you very much! I will repeat the analysis with the different power settings, but how would I know if there would be an effect on the overall results?

[Kevin Blighe]

You will see differences in the number of modules that are identified, and/or the modules to which each gene is assigned.

[Arindam Ghosh]

"generally choose the first power that passes 0.9 on the y-axis." Why 0.9?

[Kevin Blighe]

It is the value used in the tutorial by the WGCNA developer, so, it represents a useful starting point:

• Network analysis of liver expression data in female mice

[Vasantha Kumar]

We have to choose the value that is near 0.9 in the y-axis, because at that specific value is where the genes form scale free network. Barabasi et al has proved that the scale free network is biologically related and can expose hub genes. Thats the reason why we choose!!!