Dear expert:

    I wonder whether the NBS can be used to analyze the directed network? If yes, how to do it? Thanks for help.


                         Best Regards
                         Jay Chen

Hi Jay,

The NBS that is available for download does not support directed networks.

However, the NBS has been generalized based on the concept of strong and weak components in directed graphs. I have attached a script (no GUI unfortunately) that implements this directed graph version of the NBS. You might need some understanding of Matlab syntax to get it working for your data. Happy to assist you,

Andrew

Originally posted by Jay Chen:

Dear Andrew：

I appreciate your help.

Jay Chen

Hi Andrew,

I utilized your given script for the implementation of the directed graph and it works fine. Although as far as I can tell it considers only Network-Based-Statistics for the multiple comparisons. Do you have an available script that implements also (or only) FDR correction?

Thanks in advance,
Nikolas

Hi Nikolas,

In short - no. The script only implements the directed version of NBS.

Andrew

Originally posted by Nikolas Chalas:

Hi Andrew,

I was wondering whether you could assist me in understanding the output from this directed version of NBS?

I am comparing directed connectivity values between two separate groups (the first group being individuals with Alzheimer's, and the second Controls).
The contrast I am using is [1 -1]. Am I right in understanding that any significant result would suggest that there is greater connectivity in individuals with Alzheimer's compared to Controls?


The reason I ask is that the directed script outputs a variable called test_stat_mat within which I assume are the t-statistics. However, in this instance, they are all negative values. My intuition is that the negative test statistic values show that in fact they have decreased connectivity (even though the contrast is [1 -1]).


Am I misinterpreting these results? 

Best regards,
Greg

Hi Greg,

this will depend on your design matrix. If column 1 models patients and column 2 models controls, then [1 -1] will test the alternative hypotheses that control are lower than patients. The opposite hypothesis will be tested for [-1 1].

This sign of the t-statistics will depend on your design matrix and contrast. 

As a sanity check, you can also consider one of the edges in the significant network and check that the average connectivity in each groups for that edge is in the right direction.

Andrew



Originally posted by Gregory Humble:

Hi Greg,

this will depend on your
design matrix. If column 1 models patients and column 2 models controls,
then [1 -1] will test the alternative hypotheses that control are lower
than patients. The opposite hypothesis will be tested for [-1 1].

This sign of the t-statistics will depend on your design matrix and contrast. 

As
a sanity check, you can also consider one of the edges in the
significant network and check that the average connectivity in each
groups for that edge is in the right direction.

Andrew

Thanks for your assistance Andrew.

Appreciate it!

Greg