Dear all,
I'm new to PPI or gPPI and I have this question: is the task time
course not mean-centered? If so, why?
I'm asking because I'm only looking at changes in functional
connectivity during the condition and rest, or during 'on' and 'off'. It
makes sense to mean-centered the task time course, so that PPI regressor
will be correlated with seed during 'on', but anti-correlated during 'off'
(the message in figure 2 in 'Tools of the trade: psychophysiological
interactions and functional connectivity' by Dr. JX O'Reilly). Thanks a lot!

RESPONSE: PPI/gPPI is the interaction of your task and physiological
activity. In the model, you want to capture the change in coupling between
the task and the implicit baseline. To do this, you don't want to
mean-center the task timecourse. In the case of a single condition plus
rest, then it works out the same whether you center or don't center.
However, negative PPI should not be interpreted as anti-correlations -
simply a decrease in the coupling. This is supported by the fact that doing
a continuous task in the scanner does not lead to anti-correlations in the
same networks that were correlated at rest and vice versa.

In the gPPI toolbox, each the interaction regressor is detrended, but only
after the computation of the interaction. The reason for this is that the
non-task TRs should not contribute to the estimation of the coupling.

If you have multiple conditions, then you need to use gPPI and you should
not mean-center the task or seed timecourses. Each interaction represents
the change from the implicit baseline. If you have multiple conditions, but
are only interested in rest or during 'on' and 'off', then you still need
to use gPPI because your experiment has more than two conditions. As
illustrated in McLaren et al. 2012 paper on gPPI, any time you do not model
the connectivity of each task, you can get false positives and false
negatives of the two conditions you are testing.

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