Dear NBS experts,

Hello! First of all, thank you for the great analysis tool.

I have a question regarding the output of subnetwork and its raw connectivity edge strength of input.
I got a subnetwork with contrasting Control group > Disease group , but as I checked the raw edge strength of the significant subnetwork, the mean value of the strength is greater in Disease group than Control group. Which is opposite directionality from the subnetwork.
How can I explain this discrepancy?

I believed I was confident in NBS, but now I am quite confusing.
Is this a problem of my data?
Please please give me a feedback on this issue.

Thank you!

Sincerely,
Jean.

Hi Jean, 

You may want to check your contrast and design matrix. 

If your contrast is Control > Disease, all edges comprising any significant network should be consistent with this contrast (i.e. mean edge strength should be greater in Controls compared to Disease), unless confounds are having a significant impact. 

One possibility is that your design matrix includes confounds/co-variates, which have a significant impact. So, I would suggest to run without any confounds as well. 

Andrew
Originally posted by YaeJi Kim:

Dear Andrew, 

Thank you for your kind response! 
I just ran the NBS without covariates, and it shows significant subnetworks for the contrast of Disease > Control. 
However, with the covariates, it shows opposite subnetworks of Control > Disease. 
In this case, is it okay to interpret confounds have a significant impact? 

One more thing, 
 As I ran the t-test for each edge from the significant subnetworks with controlling the covariates, still some edges show opposite directionality, which is Disease > Control. 
Is this also possible, even though I controlled all the covariates? 

Thank you! 
 
Sincerely, 
Jean

Hi Jean, 

if the direction reverses with the inclusion of confounds, I think it is reasonable to assume that the confounds are having a substantial impact. 

I would check your design matrix, since it does seem a bit odd that the direction reverses with inclusion of confounds. 

Regarding your last question, all the edges in the subnetwork should show the same direction as the subnetwork as a whole. If the subnetwork shows Disease > Control, it should not be possible for edges in the subnetwork to show Disease < Control. 

I think that you should check your design matrix. 

Andrew

Originally posted by YaeJi Kim: