Dear Dr. McLaren,

I have run a gPPI with a contrast A>B and then subjected this contrast to a second-level analysis where I compared a group of HC versus patients (which was significant for HC > Pat). I would now like to interpret these results further:
- in oder to interpert this interaction can I just extract the PPI values with standard tools such as rfx Plot/marsbar?
- and how do I interpret these values? For instance, I have seen in the literature that e.g. positive values in HC vs. negative values in pat. (for contrast A>B) are sometimes interpreted as greater connectivity in HC (and no connectivity in pat) oder sometimes as positive interaction (of the regions) in HC and negative interaction in pat.

Thank you very much & best regards,

Steffi

Steffi,

Q: - in oder to interpert this interaction can I just extract the PPI
values with standard tools such as rfx Plot/marsbar?

A: I don't believe rfx/marsbar can extract the PPI values. I'd suggest
using spm_summarise to extract values from the VOI.

Q:- and how do I interpret these values? For instance, I have seen in the
literature that e.g. positive values in HC vs. negative values in pat. (for
contrast A>B) are sometimes interpreted as greater connectivity in HC (and
no connectivity in pat) oder sometimes as positive interaction (of the
regions) in HC and negative interaction in pat.

A: This means that the slope difference between A and B is higher in HC,
than in in Patients. To know if the connectivity difference of A and B in
Patients is not significantly different than 0, you'd need to test whether
they are different from 0 or not. The same is true for the healthy
controls. As with task differences, you need to know what the effect is in
each group. I would also suggest looking at the connectivity difference
between groups for task A and task B to get a more complete picture of the
connectivity. The difference of A and B, doesn't tell if A and B are both
greater in HC or if both less or if A is greater. It is important to
recognize that the caveats of condition comparisons in task data also apply
to PPI data.

Thank you very much for your reply!

I have looked now at the effects separately for Task A and B in each group in oder to understand the interaction better. Is ist meaningful if the values (i extracted them with the VOI eigenvariate button in SPM) are positive or negative, i.e. meaning positive or negative connectivity between my regions? Or if not, how could I assess positive or negative connectivity?

Steffi,

The effects of Task A and Task B separately tell you how each task behaves
relative to the implicit baseline. If one is positive then the slope of the
relationship is greater than baseline. If one is negative then the slope of
the relationship is less than baseline.

I haven't run the simulations to test if you can add the seed region beta
to the individual tasks to get an overall measure of if the connectivity is
positive or negative, but it is on my to do list. From a theoretical
perspective, it should work, but there may be some issues with the
collinearity between the task and seed regressors that could lead to poor
estimates.

I'll work on coming up with some test cases that could be used to validate
the idea of adding the task connectivity to the baseline connectivity to
get more of an absolute positive/negative connection.

Keep in mind, at rest, most areas are positively coupled unless you regress
out the GSR. It's only then do you really see the negative connectivity. I
would not be surprised that the connectivity is always positive.

When I talk about PPI I talk in terms relative to the other conditions and
relative to baseline in the same way I talk about activation in terms
relative to the other conditions and relative to baseline. We know that
there is never negative neural activity (just neural activity less than
baseline).