help > Exclude nodes from the matrix for analysis
Showing 1-4 of 4 posts
Display:
Results per page:
Dec 4, 2018  08:12 AM | Marieke van der Schaaf - Donders Institute for Brain cognition and Behaviour
Exclude nodes from the matrix for analysis
Dear NBS team, 

Thank you very much for developing this toolbox!

For my analysis I want to analyse only a subset of the edges within my matrix as testing the full matrix includes many connections that I am not interested in, which unnecessarily increases my number of comparisons.

I tried doing this by replacing all connections that I am not interested in with NaNs in the data mat-file, but the results do not seem to differ from the analysis with the full matrix. I used FDR and generally T-values > 4.2 are indicated as significant in both analyses. 

Is this the right way to do this or is there a better way to exclude connection from the analysis?

Thank you very much!

Marieke
Dec 4, 2018  09:12 PM | Andrew Zalesky
RE: Exclude nodes from the matrix for analysis
Hi Marieke,

To exclude connections from testing a priori, you should replace the connection with the value of zero in the connectivity matrices of all subjects.

In other words, if a connection has weight of zero for all subjects, it will be excluded automatically.


Remember that when using FDR, a large number of permutations must be generated (100 000+). Using only 5000 permutations is NOT valid for FDR. 

Andrew


Originally posted by Marieke van der Schaaf:
Dear NBS team, 

Thank you very much for developing this toolbox!

For my analysis I want to analyse only a subset of the edges within my matrix as testing the full matrix includes many connections that I am not interested in, which unnecessarily increases my number of comparisons.

I tried doing this by replacing all connections that I am not interested in with NaNs in the data mat-file, but the results do not seem to differ from the analysis with the full matrix. I used FDR and generally T-values > 4.2 are indicated as significant in both analyses. 

Is this the right way to do this or is there a better way to exclude connection from the analysis?

Thank you very much!

Marieke
Dec 6, 2018  06:12 AM | Marieke van der Schaaf - Donders Institute for Brain cognition and Behaviour
RE: Exclude nodes from the matrix for analysis
Dear Andrew, 

Thank you for your quick response. I'm running the analysis with 500.000 permutations, it just takes ages... Thanks for the warning.

I replaced the excluded node-values to zero, but still nothing seems to change. I attached the nbs structures for the full matrix and selected matrix in a zip file. Can you tell me if the selection is correct? Maybe I'm just missing something?

I am interested in testing connectivity from network e.g. A (7 regions) to network A, B, C and D. This should reduce the nr of comparisons to 581 (7*84 regions) compared to 3486 when using the full matrix (84x84 regions). 

For this, I kept all rows and columns of network A (see datafile attached) in the matrix and set al other edges to zero, I also tried with only keeping the rows, but that gives the same results too. I now see that e.g. all test_stats are zero for those edges that I excluded, but the size of these matrices (e.g. nbs.GLM.y or nbs.NBS.test_stat) have not changed. Is their any way I can check final number of comparisons that the program uses for correction?

Best, Marieke
Dec 7, 2018  05:12 PM | Andrew Zalesky
RE: Exclude nodes from the matrix for analysis
Hi Marieke

the size of nbs.GLM.y and nbs.NBS.test_stats will not change if values are set to zero. However, for the purpose of the NBS, the values with a t-stat of zero are effectively excluded from statistical inference. This is the same as if you were to explicitly remove the values from the connectivity matrix. Note that this will not be the case for the FDR option. For FDR, all connections will be included, irrespective of whethter they are zero or not.

I suggest adjusting your connectivity matrices by removing the rows/columns that correspond to nodes that you want to exclude.

Andrew






Originally posted by Marieke van der Schaaf:
Dear Andrew, 

Thank you for your quick response. I'm running the analysis with 500.000 permutations, it just takes ages... Thanks for the warning.

I replaced the excluded node-values to zero, but still nothing seems to change. I attached the nbs structures for the full matrix and selected matrix in a zip file. Can you tell me if the selection is correct? Maybe I'm just missing something?

I am interested in testing connectivity from network e.g. A (7 regions) to network A, B, C and D. This should reduce the nr of comparisons to 581 (7*84 regions) compared to 3486 when using the full matrix (84x84 regions). 

For this, I kept all rows and columns of network A (see datafile attached) in the matrix and set al other edges to zero, I also tried with only keeping the rows, but that gives the same results too. I now see that e.g. all test_stats are zero for those edges that I excluded, but the size of these matrices (e.g. nbs.GLM.y or nbs.NBS.test_stat) have not changed. Is their any way I can check final number of comparisons that the program uses for correction?

Best, Marieke