Hi,

I have data from an event related study in which participants had to respond to either a green or red stimuli (i.e., two conditions: Red vs Green). Additionally the participants were either individuals with autism (ASD) or typically developing (TD). Thus I have a 2 (diagnosis: ASD vs TD) x 2 (condition: Red vs Green) design.

I want to look at potential differences in functional connectivity between conditions and diagnosis group within predefined ROIs. I have run a second-level gPPI analysis with the contrasts:

Between subjects
ASD     TD
 1         -1

Within Subjects
Green     Red
  -1           1

My results show a single fronto-parietal connection as displaying an interaction between the two factors. My question is - how do I follow this up with post-hoc testing, and how do I interpret the results? I have tried two approaches:

1) In the gPPI second-level results I exported the values for red and green trials separately for ASD and TD participants and got the pattern of data in panel A of the attached image. How would I interpret these data? My guess would be that this shows that activity in these two ROIs is more associatedon red trials than green trials for ASD, but not for TD... however is there a way to tell whether that association is a positive or negative one (i.e., more associated could mean that they have a greater anti-correlation)?

2) Based on point 3 in https://www.nitrc.org/forum/message.php?msg_id=14331 I also took the same approach but with a weighted-GLM analysis and got the pattern of data in panel B of the attached image. This pattern would suggest that ASD participants have anti-correlation between these ROIs on green trials and a positive correlation on red trials.

Can you advise on which analysis makes the most sense to report (gPPI vs GLM)? Or are they complementary (i.e., analysis1 defines the interaction whereas analysis2 clarifies the type of association)?

Any help would be greatly appreciated!

Cheers

Hi Alfonso,

Any thoughts on the above?

Kind regards

Andrew

Trying this again.

To recap:

What is the difference in interpretation between condition-specific values extracted from a gPPI analysis, and those extracted from a weighted-GLM analysis?

Specifically, in the attached image, what would be the interpretation of the values from the gPPI analysis (panel A) in contrast to those for the GLM (panel B)? Or is the GLM just a natural follow up to the gPPI?

Any help really would be appreciated as I am a bit stuck here.
'
Thanks!

Hi Alex,

Generally, the values in gPPI analyses for an event-related design represent relative connectivity-change measures, comparing the connectivity during a task/stimulus-type to that during baseline. In your example, panel A values represent the increase/decrease in connectivity during the presentation of the 'red/green' stimuli (compared to that during baseline). The results there indicate that the stimulus presentation does not appear to modulate connectivity in Typically Developing subjects while it does so in Autism subjects (connectivity slightly increasing when responding to red stimuli, and slightly decreasing when responding to green stimuli).

The wGLM analyses in event-related designs are harder to interpret. While generally wGLM analyses will tell you the "absolute" connectivity during a task (instead of the connectivity compared to any baseline/reference condition), that interpretation is valid only when your task is relatively long (e.g. block designs with blocks larger than 10s), so we can meaningfully measure connectivity "during" the task. In an event-related design wGLM connectivity values are more similar to beta-series correlation values, which measure the inter-trial correlation in the BOLD responses to the stimulus, so they do not necessarily relate directly/easily to the notion of connectivity "during" rest/task that appears when those rest/task periods are considerably longer. That said, wGLM results offer meaningful/complementary information (see also a recent paper "Understanding psychophysiological interaction and its relations to beta series correlation" from Biswal group discussing some of these issues in much more detail). In your case, for example, it is interesting that panel B results show a similar pattern across groups/conditions to gPPI, indicating that the gPPI effects are likely not driven by potential differences during baseline in the connectivity between these two regions in TYP compared to ASD subjects

In any way, to further complement these results, I would probably suggest to also compute baseline connectivity in the two groups (e.g. simply create an additional "rest/baseline" condition and run wGLM analyses between these two regions during this new "rest/baseline" condition). That will give you an idea of what the absolute baseline connectivity is within each group, which can help you better frame/interpret the observed task-modulated gPPI increases/decreases in connectivity

Hope this helps
Alfonso

Originally posted by ajgordon:

Thanks Alfonso, this is really helpful!

One follow-up question: when you say the results in panel A suggest that "connectivity slightly increasing when responding to red stimuli", does that connectivity 'increase' represent an increased positive correlation between the two ROIs, or is it possible that it represents an increased negative correlation? Both possibilities would be increased connectivity, just in different directions... or is this already made clear by the wGLM analysis?

Thanks again!

Hi Alex,

That would be the information provided (more precisely than panel B) by running the additional "rest/baseline" wGLM analyses. The baseline connectivity in those analyses will be the baseline (either positive or negative) functional connectivity between the two ROIs, to which the red/green stimulus presentation adds (red) or subtracts (green) from. 

Best
Alfonso
Originally posted by ajgordon:

Thanks Alfonso, this is a big help!