Hi Alfonso,

I went forward with your suggestion and I will say that the results generally make a lot more sense. However, we are still observing this phenomenon of the largest cluster being the seed. 

We are using:

• 1st level gPPI selection: task-dependent total-connectivity


• -1 1; contrast between our two conditions


• Block design

In the attached file, you can see an example of this—we used the IPS (R) as a seed, and the biggest cluster by far was simply the IPS. It observed a positive effect in our second conditions (corresponding to the "1" in the contrast) compared to our first condition (corresponding to the "-1" in the contrast)—this seems to be the case every time. I can't help but think something is still going on, even though the problem has improved somewhat, with the cluster being much smaller than before.

1. Are there any other settings that may be creating this issue?
   
2. If not, how might we interpret this phenomenon? 
   

I really appreciate your help—any advice would be amazing.

Originally posted by Alfonso Nieto-Castanon:

Hi,

Can you please let me know whether when defining your first-level gPPI analysis you are selecting the option that reads "gPPI model output: interaction coefficients (task-dependent connectivity-change)" or the one that reads "gPPI model output: main+interaction coefficients (task-dependent total-connectivity)" (the exact text may be slightly different in prevoius versions of CONN)? and whether your design was an event-related or a block- design?

My guess would be that you may have the second option selected (main+interaction) which, combined with your "any effects" (F-test, 1,0;0,1) between-tasks contrast, is making your second-level analysis evaluate in what areas there is significant connectivity with the seed in either of your two tasks (or both). If instead you would like to evaluate in what areas there is a significantly different connectivity with the seed between the two tasks you can do that using the "difference" ( [-1 1] ) between-tasks contrast.  The specific options and interpretation would change a bit depending on whether you have an event-related design where you often want to explore difference between each task and the design implicit baseline (when no stimuli/events were presented) vs. when you have a block design where any experimental baseline is often defined explicitly as a separate condition. 

Hope this helps

Alfonso

Originally posted by sfeuer45:

Hello all! 

I am doing some research related to task-based connectivity. We are using CONN's implementation of gPPI, and then performing GLM analysis to compare connectivity across participants between tasks. I've noticed that no matter what region I use as a seed and search for any effects (F-test, contrast 1,0;0,1;) between tasks, there is a huge cluster that is called significant by the GLM analysis that includes that region. It also often includes the other regions in the network containing the selected seed. 

For instance, in the attached file, we use the ACC within the salience network as a seed, and the results include a large cluster that combines the ACC with the rest of the salience network. 

I have a few questions about this that partially relate to my general unfamiliarity with the mathematics behind GLMs:

• Should we expect this to happen every time? (If not, am I using the model incorrectly?)


• Why does this happen / what does it mean for the connectivity of a seed region to vary with itself between tasks?

Please let me know if you need any more information to help—any information would be appreciated, as I've been thinking about this issue for multiple weeks.