Dear Bioconductor users,

I am using WGCNA software for RNA-seq data (320 patient samples) and I want to decipher potential mechanisms underlying hepatocellular carcinoma progression. First, I removed all features that have a count of less than 10 in more than 90% of the samples and I transformed my data using varianceStabilizingTransformation function implemented in DEseq2 package. Second, I removed all potential outliers and then I used the pickSoftThreshold function to select  the beta power for signed network construction. As you can see in the attached plot, the SFT fit index of power of 12 (default option) fails to reach the 0.8. The power of 16 fits the Scale free topology. Although the power of 16 means network modules with low mean connectivity. Which is your opinion about the selection of soft threshold power in my case?  Is it absolutely necessary the selection of the soft threshold power based in SFT criterion?

https://www.dropbox.com/s/yyb3a4houef2utl/SFT_MEAN_CONN.pdf?dl=0

Thank you for your time in advance!!

Sincerely,

Panagiotis Mokos

Please read the WGCNA FAQ, particularly point 6. You can use the default power  of 12 since you have hundreds of samples, but do make sure you don't have strong uninteresting sources of variation as described in the FAQ.

Looking at your plot I suspect you have several subgroups which affect your analysis.  When you do a PCA or a MDS of the 320 samples (after normalization) do they form a single group?

If you want to see progression of the hepatocellular carcinoma you should divide into several groups for "phases" and see the co-expression in each phase. (If there isn't any clinical description of the phases of your samples you could use the PCA to select the phases)

To know the expression which is constantly co-expressed in the progression a consensus network and a preservation analysis to know which modules are preserved against a reference (I don't know what could be a reference, maybe an early tumor).

Dear Langfelder and Lluis,

Thank you very much for your quick replies!!!

I have constructed the cluster dendrogram (including the trait heatmap) of samples based on their Euclidean distance, according to the specifications of WGCNA tutorial, as you can see in the attached plot. Specifically, white color means low grade/ early stage, and red indicates high grade/ late stage. Do you believe that my data have subgroups which affect scale-free criterion (sft) fit? 

https://www.dropbox.com/s/e9njnylivhy9cn6/SAMPLE_DENDROGRAM_AND_TRAIT_HEATMAP.pdf?dl=0

I would appreciate hearing your opinion!!

Thank you very much for your time in advance!!

Sincerely,

Panagiotis Mokos