help
help > RE: Correlating graph measurements
Oct 8, 2016 06:10 AM | Pravesh Parekh - University of Oslo
RE: Correlating graph measurements
Hi Aser,
I can partially respond to your questions.
1. Yes, this is the usual method for defining contrasts. You can simply select the group which you want (if its just one group) and check the graph theory results. When you need to compare these two, set up a contrast (for example 1 -1) and the graph theory results would be of that particular contrast. Do note that because you are interested in specific defined network, select appropriate ROIs only before opening graph theory results (applicable in case you have more ROIs).
The results you see in the graph theory results window are mean across subjects. You can access individual subject's values as well (see below), if that is what you are looking for. Just to clarify here: the global efficiency that you see is the group mean global efficiency. Conn would calculate global efficiency for each subject, then depending on your contrast, it might do a t test to see if the mean global efficiency is significantly non-zero, then display that mean value (or the difference in mean value) in the results pane along with p value.
I guess comparing global efficiency (or a different graph theory measure) of two different data for the same set of subjects will show some interesting features. For example, a condition like schizophrenia with known network integration deficits, it might be of interest to see how different functional integration parameters differ when performing a cognitive task and when at rest, to see if there are some networks which are impaired only in task, etc.
I have not implemented gPPI in Conn myself but as long as you have a measure of functional connectivity between a pair of ROIs (nodes), you can (in theory) calculate graph theory parameters by considering those ROIs as nodes and the functional connectivity measure as the edge connecting those nodes. I would think that Conn would give you the graph theory results for gPPI too (but again, I have not checked this).
Regarding your P.S, you can export your adjacency matrix from the GUI. This would give you subject level adjacency matrices and you can calculate any graph theory measure that you would like for individual subjects (if you need that, as above). You should also be able to access some of these from the second level results folder in your conn project folder (I am away from lab when writing this so cannot check the exact file which might be of interest). The adjacency matrix can be used to get the desired result in the matrix form, as you need (in case the results folder does not already have that file stored). Also, in the conn graph theory GUI, you can edit out the threshold value and press enter (i.e. press enter after making it blank) and it would plot a quick graph of the measure varying across thresholds.
Hope the above helps.
Best
Pravesh
Originally posted by Aser A:
I can partially respond to your questions.
1. Yes, this is the usual method for defining contrasts. You can simply select the group which you want (if its just one group) and check the graph theory results. When you need to compare these two, set up a contrast (for example 1 -1) and the graph theory results would be of that particular contrast. Do note that because you are interested in specific defined network, select appropriate ROIs only before opening graph theory results (applicable in case you have more ROIs).
The results you see in the graph theory results window are mean across subjects. You can access individual subject's values as well (see below), if that is what you are looking for. Just to clarify here: the global efficiency that you see is the group mean global efficiency. Conn would calculate global efficiency for each subject, then depending on your contrast, it might do a t test to see if the mean global efficiency is significantly non-zero, then display that mean value (or the difference in mean value) in the results pane along with p value.
I guess comparing global efficiency (or a different graph theory measure) of two different data for the same set of subjects will show some interesting features. For example, a condition like schizophrenia with known network integration deficits, it might be of interest to see how different functional integration parameters differ when performing a cognitive task and when at rest, to see if there are some networks which are impaired only in task, etc.
I have not implemented gPPI in Conn myself but as long as you have a measure of functional connectivity between a pair of ROIs (nodes), you can (in theory) calculate graph theory parameters by considering those ROIs as nodes and the functional connectivity measure as the edge connecting those nodes. I would think that Conn would give you the graph theory results for gPPI too (but again, I have not checked this).
Regarding your P.S, you can export your adjacency matrix from the GUI. This would give you subject level adjacency matrices and you can calculate any graph theory measure that you would like for individual subjects (if you need that, as above). You should also be able to access some of these from the second level results folder in your conn project folder (I am away from lab when writing this so cannot check the exact file which might be of interest). The adjacency matrix can be used to get the desired result in the matrix form, as you need (in case the results folder does not already have that file stored). Also, in the conn graph theory GUI, you can edit out the threshold value and press enter (i.e. press enter after making it blank) and it would plot a quick graph of the measure varying across thresholds.
Hope the above helps.
Best
Pravesh
Originally posted by Aser A:
Dear Alfonso,
I have two groups and for one of the group I have clinical data. I would like to first extract and look at the global efficiency of a specific defined network between the two groups ? How can I perform this ? I think I tried something like first: contrasting each group separately and press the graph theory button. And if I want to compare the two group I contrast them like 1 -1 and press the graph theory button. Is this correct ? Any comment is appreciate it here ?
Another important comment which I could not find an answer to is that I would like to correlate the (e.g. global efficiency with the clinical data). The clinical data is a value per subject. How can I look at this ?
Lastly, Does it make sense if I compare (e.g. global efficiency) of two different data (for example resting state vs task data) or (2 different task data). Also does conn compute the graph measurements for task data of gPPI.
Many thanks
Aser
I have two groups and for one of the group I have clinical data. I would like to first extract and look at the global efficiency of a specific defined network between the two groups ? How can I perform this ? I think I tried something like first: contrasting each group separately and press the graph theory button. And if I want to compare the two group I contrast them like 1 -1 and press the graph theory button. Is this correct ? Any comment is appreciate it here ?
Another important comment which I could not find an answer to is that I would like to correlate the (e.g. global efficiency with the clinical data). The clinical data is a value per subject. How can I look at this ?
Lastly, Does it make sense if I compare (e.g. global efficiency) of two different data (for example resting state vs task data) or (2 different task data). Also does conn compute the graph measurements for task data of gPPI.
Many thanks
Aser
Threaded View
| Title | Author | Date |
|---|---|---|
| Aser A | Oct 7, 2016 | |
| Pravesh Parekh | Oct 8, 2016 | |
| Aser A | Oct 13, 2016 | |
| Aser A | Oct 7, 2016 | |