help > NBS for comparing networks within subjects but different edges
Showing 1-3 of 3 posts
Oct 7, 2022 09:10 AM | Marcela Ovando-Tellez
NBS for comparing networks within subjects but different edges
Dear profesor Zalesky,
First of all, thank you very much for this wonderful toolbox and the recent one, NBS-predict.
I have been exploring the NBS toolbox since I have the following issue:
I have the data of 93 subjects. I only have one set of fMRI data for each subject and I used a CPM approach to predict two different behavioral data.
From these CPM analyses I obtained a 'mask' with the significant edges to predict behavior 1 (82 edges) , and another one for the prediction of behavior 2
(28 edges). However, these two masks have some edges in common (18 in total).
Now, I want to apply some statistical inference on those connections differentiating between behavior 1 and behavior 2.
Using the NBS toolbox, I created a paired t-test design matrix by using the weighted networks for each participant that predict behavior 1 (condition 1), and the one predicting behavior 2 (condition 2).
I defined the parameters as follows:
Design matrix (example with 4 subjects):
1 0 0 0 1
0 1 0 0 1
0 0 1 0 1
0 0 0 1 1
1 0 0 0 -1
0 1 0 0 -1
0 0 1 0 -1
0 0 0 1 -1
The Matrices are organized as follows: the weighted matrices that predict behavior 1 for participants 1 to 93, followed by the weighted matrices that predict behavior 2 for participants 1 to 93.
The exchanged blocks were defined as [1:93,1:93]
However, since the fMRI data is the same for each participant, and the only difference between the 'condition 1' and the 'condition 2' are the weight of the selected edges based on the CPMs, I am not sure if I am using NBS method in an appropriate way. If the approach is correct, do you recommend to use NBS or the FDR method?
Thank you very much in advance.
Marcela Ovando-Tellez
First of all, thank you very much for this wonderful toolbox and the recent one, NBS-predict.
I have been exploring the NBS toolbox since I have the following issue:
I have the data of 93 subjects. I only have one set of fMRI data for each subject and I used a CPM approach to predict two different behavioral data.
From these CPM analyses I obtained a 'mask' with the significant edges to predict behavior 1 (82 edges) , and another one for the prediction of behavior 2
(28 edges). However, these two masks have some edges in common (18 in total).
Now, I want to apply some statistical inference on those connections differentiating between behavior 1 and behavior 2.
Using the NBS toolbox, I created a paired t-test design matrix by using the weighted networks for each participant that predict behavior 1 (condition 1), and the one predicting behavior 2 (condition 2).
I defined the parameters as follows:
Design matrix (example with 4 subjects):
1 0 0 0 1
0 1 0 0 1
0 0 1 0 1
0 0 0 1 1
1 0 0 0 -1
0 1 0 0 -1
0 0 1 0 -1
0 0 0 1 -1
The Matrices are organized as follows: the weighted matrices that predict behavior 1 for participants 1 to 93, followed by the weighted matrices that predict behavior 2 for participants 1 to 93.
The exchanged blocks were defined as [1:93,1:93]
However, since the fMRI data is the same for each participant, and the only difference between the 'condition 1' and the 'condition 2' are the weight of the selected edges based on the CPMs, I am not sure if I am using NBS method in an appropriate way. If the approach is correct, do you recommend to use NBS or the FDR method?
Thank you very much in advance.
Marcela Ovando-Tellez
Oct 8, 2022 05:10 AM | Andrew Zalesky
RE: NBS for comparing networks within subjects but different edges
Hi Marcela,
the design matrix and other details seem ok. However, I am not sure if NBS or FDR is really the right kind of tool for your aims here.
If you are simply wanting to test whether the overlap in edges (18) is statistically significant, I think that there are simpler approaches that you could consider.
For example, you could randomize the behavioral data between the two conditions, re-run the CPM on the randomized data and store the number of edges that overlap. This can be repeated for different randomizations to build an empirical null distribution for the overlap in edges. It will be important to ensure that the number of significant edges is the same for each condition in the randomize data (not sure if this is possible with CPM).
If you have X significant connections in condition 1 and Y significant connections in condition 2, and even simpler approach would be to randomly select X connections and then randomly select Y connections form anywhere in the brain network. The overlap can then be computed to build up a null distribution.
I hope this helps.
Best,
Andrew
Originally posted by Marcela Ovando-Tellez:
the design matrix and other details seem ok. However, I am not sure if NBS or FDR is really the right kind of tool for your aims here.
If you are simply wanting to test whether the overlap in edges (18) is statistically significant, I think that there are simpler approaches that you could consider.
For example, you could randomize the behavioral data between the two conditions, re-run the CPM on the randomized data and store the number of edges that overlap. This can be repeated for different randomizations to build an empirical null distribution for the overlap in edges. It will be important to ensure that the number of significant edges is the same for each condition in the randomize data (not sure if this is possible with CPM).
If you have X significant connections in condition 1 and Y significant connections in condition 2, and even simpler approach would be to randomly select X connections and then randomly select Y connections form anywhere in the brain network. The overlap can then be computed to build up a null distribution.
I hope this helps.
Best,
Andrew
Originally posted by Marcela Ovando-Tellez:
Dear profesor Zalesky,
First of all, thank you very much for this wonderful toolbox and the recent one, NBS-predict.
I have been exploring the NBS toolbox since I have the following issue:
I have the data of 93 subjects. I only have one set of fMRI data for each subject and I used a CPM approach to predict two different behavioral data.
From these CPM analyses I obtained a 'mask' with the significant edges to predict behavior 1 (82 edges) , and another one for the prediction of behavior 2
(28 edges). However, these two masks have some edges in common (18 in total).
Now, I want to apply some statistical inference on those connections differentiating between behavior 1 and behavior 2.
Using the NBS toolbox, I created a paired t-test design matrix by using the weighted networks for each participant that predict behavior 1 (condition 1), and the one predicting behavior 2 (condition 2).
I defined the parameters as follows:
Design matrix (example with 4 subjects):
1 0 0 0 1
0 1 0 0 1
0 0 1 0 1
0 0 0 1 1
1 0 0 0 -1
0 1 0 0 -1
0 0 1 0 -1
0 0 0 1 -1
The Matrices are organized as follows: the weighted matrices that predict behavior 1 for participants 1 to 93, followed by the weighted matrices that predict behavior 2 for participants 1 to 93.
The exchanged blocks were defined as [1:93,1:93]
However, since the fMRI data is the same for each participant, and the only difference between the 'condition 1' and the 'condition 2' are the weight of the selected edges based on the CPMs, I am not sure if I am using NBS method in an appropriate way. If the approach is correct, do you recommend to use NBS or the FDR method?
Thank you very much in advance.
Marcela Ovando-Tellez
First of all, thank you very much for this wonderful toolbox and the recent one, NBS-predict.
I have been exploring the NBS toolbox since I have the following issue:
I have the data of 93 subjects. I only have one set of fMRI data for each subject and I used a CPM approach to predict two different behavioral data.
From these CPM analyses I obtained a 'mask' with the significant edges to predict behavior 1 (82 edges) , and another one for the prediction of behavior 2
(28 edges). However, these two masks have some edges in common (18 in total).
Now, I want to apply some statistical inference on those connections differentiating between behavior 1 and behavior 2.
Using the NBS toolbox, I created a paired t-test design matrix by using the weighted networks for each participant that predict behavior 1 (condition 1), and the one predicting behavior 2 (condition 2).
I defined the parameters as follows:
Design matrix (example with 4 subjects):
1 0 0 0 1
0 1 0 0 1
0 0 1 0 1
0 0 0 1 1
1 0 0 0 -1
0 1 0 0 -1
0 0 1 0 -1
0 0 0 1 -1
The Matrices are organized as follows: the weighted matrices that predict behavior 1 for participants 1 to 93, followed by the weighted matrices that predict behavior 2 for participants 1 to 93.
The exchanged blocks were defined as [1:93,1:93]
However, since the fMRI data is the same for each participant, and the only difference between the 'condition 1' and the 'condition 2' are the weight of the selected edges based on the CPMs, I am not sure if I am using NBS method in an appropriate way. If the approach is correct, do you recommend to use NBS or the FDR method?
Thank you very much in advance.
Marcela Ovando-Tellez
Oct 10, 2022 04:10 PM | Marcela Ovando-Tellez
RE: NBS for comparing networks within subjects but different edges
Dear professor,
thank you very much for your help and quick response.
I was almost sure that the NBS was not the more appropriate method for my aim, therefore, thanks for the confirmation. I will try what you proposed.
Best regards,
Marcela Ovando-Tellez
thank you very much for your help and quick response.
I was almost sure that the NBS was not the more appropriate method for my aim, therefore, thanks for the confirmation. I will try what you proposed.
Best regards,
Marcela Ovando-Tellez