What does the node size in the network() output represent?

Analysis
1

Dear mixOmics community,

I am new to your package and want to congratulate you on this easily accessible bioinformatical tool box!
I ran an rCCA model (N = 38) using the shrinkage method after NIPALS-imputation on a microbial data set (206 variables) and a sample data set (25 variables) - however, I have two questions that I was not able to answer yet:

  1. when running the network() function, the output generates differently sized nodes (annotated as “size” when importing to Cytoscape). What does the node size represent?
  2. I have tested the overall model for signficance by extracting the canonical variates and performaing a spearman calculation via cor.test(), yielding a p-value of < 2.2e-16. Is there a method for testing each canonical correlation for significance? I have tried using the p.asym() function of the CCP package, but this only returns NaNs as p-values.

Thanks a lot for your help in advance!

Best,
Luise

2

hi @l.bel,

  1. when running the network() function, the output generates differently sized nodes (annotated as “size” when importing to Cytoscape). What does the node size represent?

Nothing :slight_smile: it is coming from the iGraph package and the size of the node depends on the length of the variable name.

  1. I have tested the overall model for signficance by extracting the canonical variates and performaing a spearman calculation via cor.test(), yielding a p-value of < 2.2e-16. Is there a method for testing each canonical correlation for significance? I have tried using the p.asym()function of the CCP package, but this only returns NaNs as p-values.

rCCA (assuming you are using a regularisation parameter given than P+Q >> N) will always lead very high correlations no matter the test you use. It’s best to either look at whether the graphical outputs make sense (i.e sample plots - do the sample cluster together according to your biological assumption; variable plots - do the clusters of bacteria and sample make sense?). rCCA also does not perform variable selection so it become very trick to make sense of the results. You should consider a sparse sPLS canonical mode, the outputs are similar, and you can calculate the correlation between variates. You can have variable selection in this method to extract interesting associations between bacteria and samples.

Kim-Anh

3

Hi Kim-Anh,
thank you so much for your quick help - i will follow your suggenstion and code an sPLS model :slight_smile:
have a great day!
best,
Luise

1 Like