help
help > RE: NBS one sample-test with more negative than positive values/thresholding of connectivity matrices.
Mar 3, 2022 08:03 AM | Andrew Zalesky
RE: NBS one sample-test with more negative than positive values/thresholding of connectivity matrices.
Hi Ole,
Regarding my previous comment - in general it is not necessary to check that all the values are positive. I suggested this sanity check to assess whether there was a definite effect in the person's data (i.e. null rejection). It is more likely for a significant effect to emerge if all connections are positive. Please don't look too much into this comment. It was simply a suggestion to get a basic idea about whether there was an effect in the data. But it is definitely not necessary to check the polarity of connections before running NBS.
Yes - thresholding is a delicate issue and can introduce numerous confounds. You might want to check out this paper for some of the issues associated with proportional thresholding in case-control studies: https://pubmed.ncbi.nlm.nih.gov/28167349/
Apologies if my response is not helpful, but these issues are not easy to discuss via email.
best wishes, Andrew
Originally posted by Ole Jonas Böken:
Regarding my previous comment - in general it is not necessary to check that all the values are positive. I suggested this sanity check to assess whether there was a definite effect in the person's data (i.e. null rejection). It is more likely for a significant effect to emerge if all connections are positive. Please don't look too much into this comment. It was simply a suggestion to get a basic idea about whether there was an effect in the data. But it is definitely not necessary to check the polarity of connections before running NBS.
Yes - thresholding is a delicate issue and can introduce numerous confounds. You might want to check out this paper for some of the issues associated with proportional thresholding in case-control studies: https://pubmed.ncbi.nlm.nih.gov/28167349/
Apologies if my response is not helpful, but these issues are not easy to discuss via email.
best wishes, Andrew
Originally posted by Ole Jonas Böken:
Hi Andrew,
Thanks again for your quick reply!
I will try to clarify:
1.) Yes, having larger random components means that the actual component is not significant and we are aware that this represents not a general problem.
However, since we do not find any results (i.e., significant components) in our data, we suspect that this might be caused by a mismatch between negative (>) positive connections. Because this mismatch leads to having larger random than actual components while permuting the data with the NBS.
Why?:
„If this does not work, check that your connectivity values are indeed positive for the majority of samples."
Found here: https://www.nitrc.org/forum/forum.php?th...
Maybe it would help us to better understand, if you could explain what you meant with that post and why someone should check that connectivity values are positive for the majority of samples and if/how this influences the NBS.
2.) Our question regarding the thresholding was not whether a mismatch would preclude the use of a thresholding. It's the other way around, we are asking ourselves, if such an imbalance would justify the use of such a thresholding.
Hope I was able to clarify.
Best,
Ole
Thanks again for your quick reply!
I will try to clarify:
1.) Yes, having larger random components means that the actual component is not significant and we are aware that this represents not a general problem.
However, since we do not find any results (i.e., significant components) in our data, we suspect that this might be caused by a mismatch between negative (>) positive connections. Because this mismatch leads to having larger random than actual components while permuting the data with the NBS.
Why?:
- As I stated in my previous post. If we apply light proportional thresholding to erase (e.g., 5% of) negative connections, we get results that we deem meaningful.
- You also stated that:
„If this does not work, check that your connectivity values are indeed positive for the majority of samples."
Found here: https://www.nitrc.org/forum/forum.php?th...
Maybe it would help us to better understand, if you could explain what you meant with that post and why someone should check that connectivity values are positive for the majority of samples and if/how this influences the NBS.
2.) Our question regarding the thresholding was not whether a mismatch would preclude the use of a thresholding. It's the other way around, we are asking ourselves, if such an imbalance would justify the use of such a thresholding.
Hope I was able to clarify.
Best,
Ole
Threaded View
| Title | Author | Date |
|---|---|---|
| Ole Jonas Böken | Feb 17, 2022 | |
| Ole Jonas Böken | Feb 28, 2022 | |
| Andrew Zalesky | Mar 3, 2022 | |
| Ole Jonas Böken | Feb 23, 2022 | |
| Andrew Zalesky | Feb 24, 2022 | |
| Andrew Zalesky | Feb 17, 2022 | |