Dear Dr. McLaren

I've been using the gPPI toolbox as well as GLMFlex for some time now, and I would like to thank you for these really useful toolboxes.

I am currently using the gPPI toolbox on my parametric factorial design experiment to examine whether there were negative interactions associated with a decrease in the S2 seed region with one of my factors. However, the gPPI showed a negative interaction within the seed region. I tried multiple variations of PPIs (see an example below), including creating the PPI regressors manually and all gave me the same results. It is not clear to me if there is a fundamental problem with the PPI, or if there is some meaning behind the results. Any insight you have on this would be greatly appreciated.

P.subject=['PPI_VOI'];
P.directory=sub_path;
P.outdir=outpath_PPI;
P.VOI= 'VOI.img';
P.Region=Roi_tags{roi_n};
P.Estimate=1;
P.contrast=2;
P.extract='eig';
P.Tasks={'1' 'Task'};
P.Weights=[];
P.analysis='psy';
P.method='cond';
P.CompContrasts=1;
%Task contrast
P.Contrasts(1).left={'Task'};
P.Contrasts(1).right={'none'};
P.Contrasts(1).STAT='T';
P.Contrasts(1).name='Task_name';
%Parametric modulator A contrast
P.Contrasts(2).left={'Task'};
P.Contrasts(2).right={'none'};
P.Contrasts(2).STAT='T';
P.Contrasts(2).Contrail={'xPM_A^1'};
P.Contrasts(2).name='PM_A';
%Parametric modulator B contrast
P.Contrasts(3).left={'Task'};
P.Contrasts(3).right={'none'};
P.Contrasts(3).STAT='T';
P.Contrasts(3).Contrail={'xPM_B^1'};
P.Contrasts(3).name='PM_B';


Best, 

Julio Duenas

I don't believe this is an artifact of the method.
Does the entire seed have a negative value for the PPI term?
Does it depend on the seed size?


If we start off with the resting state, we measure 1 value of connectivity over the entire resting state run. However, we know that the connectivity is likely to vary over time. Thus, it is reasonable to think that there could be some variability in the local connectivity.

If I generate 10 voxels that are perfectly connected during period 1 and then add noise to them at period 2, then I have a rough simulation of a seed region that has a change in connectivity over time. If I put these values into a PPI model with 3 columns (task, seed, task*seed), then I get a negative PPI value for the task period if the task period is period 1. Note that the connectivity isn't negative, its just less than baseline. One interpretation is that the activity is driven by an external stimuli rather than the seed as you are partially out the activity via the PSYCH regressor.

I will need to more simulations (with deconvolution) to fully understand what is happening and find physiological examples. I'll add this to my gPPI methodological analyses list.

A few possible sources of artifacts:
(1) You didn't properly model all the connectivity states. You should have 1 PPI regressor for each task.
(2) You didn't properly model the duration of the interaction. Event-related designs should not use an event duration of 0.
(3) There is an artifact in the data related to the experimental design and timing (e.g. stimulus correlated motion).

Hope this helps. Let me know if you have additional questions.