I would like to look at the connectivity within one patient group.  I tried using the one-sample button, with a design matrix of all ones and a contrast matrix of [1], with the eventual goal of adding a regressor in a design matrix. 

But I get non-signficant results for all thresholds I tried, and max size random and max size actual are both zero. 


Can I do a group mean with NBS? 
and if so, how do I set things up. 

Thanks

Hi Teena,

It's most likely that a one-sample test is not appropriate in your case, but you will need to provide more details. Are you comparing the patients to a group of controls? What is the null hypothesis that you are wanting to test?

Andrew

Originally posted by Teena Moody:

Dear Teena and Prof. Zalesky;

First things first, thanks for this usefull toolbox! Hope you can help me with some questions...
I think Teena and i are facing similar issues, so i will try to elaborate on my problem and probably you may help both..

I ´ve calculated structural connectivity matrices for a reduced group of normal subjects (n=22), and also individual matrices of (n=21) individual epilepsy patients; were each patient is different from each other.
We would like to test the null hypothesis that each individual epilepsy patient has no connectivity difference compared against our normal sample of 21 subjects.

In other words, is it possible to avoid a group to group comparation in order to test for differences on each single epilepsy patient compared against the control group (n=22) using the NBS toolbox ? In the case it is true, what kind of test should we use and how could the matrix be designed to work properly ?

Hope i was clear enough
Thanks for your time and help

Juan Pablo

Hi Juan,

You can't really test a null hypothesis about a single observation (e.g. a single individual with epilepsy compared to a group of controls). This is not something that is specific to the NBS, but rather general to statistics.

You can only test whether the patient group as a whole has no connectivity differences compared to the controls.

If you are interested about making inference at the level of an individual subject, you might want to consider predicting patients versus controls using something like a support vector machine.

Andrew

Originally posted by Juan Pablo Princich:

Hi Juan,

I would like to clarify my previous response.

If you want to compare single individual to the sample of controls, you could use a design matrix of the form:

1 1
1 0
1 0
1 0
1 0
1 0

where the first row is the individual with epilepsy and the other rows correspond to the healthy controls.

Selecting a t-test with the contrast [0 1] will test whether the connectivity in the epilepsy individual is larger than controls. The contrast [0 -1] will test the opposite.

Andrew


Originally posted by Andrew Zalesky:

Hi 
Thanks for the reply.  I really want to know what the one sample button is for. 

I have had success doing t-tests between groups, for example, looking at the NBS results for CON vs PATIENT groups using the contrasts [1,-1] and [-1,1], and for anovas.  

Can NBS also be used to identify the "strongest" networks within a single group?  I thought perhaps that was what the one sample button was for.  I couldn't find this option mentioned in the documentation. 

Also is the pair-wise option available?

Thank you!

Hi Teena Moody,

Regarding the one-sample test, here is the relevant text from the NBS manual:

"The one sample test randomly flips the sign of each data point for each permutation, which corresponds to a one-sided, one-sample t-test. Note that the one sample test always assesses whether the mean is greater than zero, irrespective of the sign of the contrast vector."


Andrew

Originally posted by Teena Moody:

Thanks Professor for the further respose!

Will 5000 permutations and 0.05 significane value be appropiate to test this experiment design?
What is your opinion , will be of interest to test the analysisi using FDR or SPC methods in this particular escenario one patient against a control group ?


Will post results as soon as possible
Thanks again for your time

N=5000 and p<0.05 sounds reasonable to me.

However, as I mentioned before, I don't have any experience in dealing with a sample size of 1.

SPC is not implemented in the current version of the software.

Andrew
Originally posted by Juan Pablo Princich:

Thanks for the information, I should have read the documentation more carefully. 
best
Teena Moody

Dear Proffesor, i followed your recomendations and some of the patients showed connectivity differences.
Now i will try to compare both groups as a hole, but would like to have your advice on how to incorporate a clinical variable for a correlation analysis and how the contrast should be defined.
For instance, i have this design and wants to run a t- test
0 1
0 1
0 1
1 0
1 0
1 0

Will be ok to add a new column directly with the clinical values for each datapoint ? And then how can the contrast be defined ?

Thanks again for your time and help
Juan

Adding covariates is discussed in the references manual. In short, simply add a new column containing the clinical variable. For example,your design matrix might look like this:

1 0.2
1 0.3
1 0.1
1 0.9
1 0.4
1 0.5

and the contrast [0 1] will test for a correlation between the 2nd column (clinical variable) and your connectivity values.

Originally posted by Juan Pablo Princich:

Thanks for taking the time to explain the details.
Just to clarify, will this design test for connectivity differences between two groups (patients against controls) considering the clinical covariates?  or all the subjects will be tested as one group in the regression analysis ?

I will do the latter - all tested as one group.

The former is described in detail in the reference manual.

Originally posted by Juan Pablo Princich: