help > Interpretation of gPPI results
Showing 1-2 of 2 posts
Feb 10, 2017 01:02 PM | Zane Xie - UCR
Interpretation of gPPI results
Dear PPI experts
I am hoping to get your help for the interpretation of gPPI results. Currently, I have a dataset with 4 conditions (within-subject). I have a very specific hypothesis on how a brain region (e.g., hippocampus) responses to these 4 conditions (monotonically increase from condition 1 to condition 4, namely control, low, medium, and high).
I ran a tradition activation-based analysis with a linear contrast for these 4 conditions and identified three major brain regions, namely mPFC, hippocampus, and angular gyrus, which followed this contrast. I then wanted to figure out whether the experimental conditions also modulated the communication across these different brain regions. Thus, I conducted a gPPI analysis. I got 4 interaction terms, and I conducted a linear contrast on the beta values of these 4 interaction terms.
I believe, the outcome of this contrast would tell me whether the connectivity between the seed region (e.g., hippocampus) and a target region also change as a linear function of my experimental manipulations. Surprisingly, I did not find any target regions that had shown up in the activation-based analysis, but found another region, the caudate, which had failed to reach the statistical threshold in the activation analysis, to positively correlated with the hippocampus. How should I interpret this PPI results in comparison to the activation results?
I learn from a tutorial that the PPI effect is independent of the typical effect (e.g., the contrast between conditions A and B). In other words, positive PPI effect has nothing to do with the sign of the contrast between conditions A and B. However, wouldn't we expect PPI analysis should also yield regions that show up in the activation-based analysis? If a certain brain region does not engage in a task to begin with (or at least fail to reach statistical significance based on some criteria, in this case the caudate), how could we justify the PPI finding of this region for a context-dependent correlation with the seed region (e.g., hippocampus)? Alternatively, since we have the taken into account all the original condition effects in our regression model for PPI analysis, could we just simply treat the PPI results as a completely independent analysis of the data to allow the detection of more subtle effect?
In sum, what the relationship between the PPI analysis and traditional activation analysis? Should one (PPI) contingent on the other (activation)? How to interpret their findings when they differ?
Thank you for your advice and clarification in advance!
Best regards,
Zane
I am hoping to get your help for the interpretation of gPPI results. Currently, I have a dataset with 4 conditions (within-subject). I have a very specific hypothesis on how a brain region (e.g., hippocampus) responses to these 4 conditions (monotonically increase from condition 1 to condition 4, namely control, low, medium, and high).
I ran a tradition activation-based analysis with a linear contrast for these 4 conditions and identified three major brain regions, namely mPFC, hippocampus, and angular gyrus, which followed this contrast. I then wanted to figure out whether the experimental conditions also modulated the communication across these different brain regions. Thus, I conducted a gPPI analysis. I got 4 interaction terms, and I conducted a linear contrast on the beta values of these 4 interaction terms.
I believe, the outcome of this contrast would tell me whether the connectivity between the seed region (e.g., hippocampus) and a target region also change as a linear function of my experimental manipulations. Surprisingly, I did not find any target regions that had shown up in the activation-based analysis, but found another region, the caudate, which had failed to reach the statistical threshold in the activation analysis, to positively correlated with the hippocampus. How should I interpret this PPI results in comparison to the activation results?
I learn from a tutorial that the PPI effect is independent of the typical effect (e.g., the contrast between conditions A and B). In other words, positive PPI effect has nothing to do with the sign of the contrast between conditions A and B. However, wouldn't we expect PPI analysis should also yield regions that show up in the activation-based analysis? If a certain brain region does not engage in a task to begin with (or at least fail to reach statistical significance based on some criteria, in this case the caudate), how could we justify the PPI finding of this region for a context-dependent correlation with the seed region (e.g., hippocampus)? Alternatively, since we have the taken into account all the original condition effects in our regression model for PPI analysis, could we just simply treat the PPI results as a completely independent analysis of the data to allow the detection of more subtle effect?
In sum, what the relationship between the PPI analysis and traditional activation analysis? Should one (PPI) contingent on the other (activation)? How to interpret their findings when they differ?
Thank you for your advice and clarification in advance!
Best regards,
Zane
Apr 13, 2017 01:04 AM | Donald McLaren
RE: Interpretation of gPPI results
(1) Connectivity change don't need to occur in regions that have
activity change. The caudate could have a very constant response on
average across all 4 trial types, but there could be modulation of
the trial to trial variability that changes with trial type and
hippocampal signal.
(2) An activated region is a region where the signal changes similarly across all trials of each trial type according to the HRF. A connected region is a region where the signal changes similarly across all trials of each trial type according to the neural activity in the seed region convolved with the HRF. As you can see, these are two fundamentally different signals. They can occur together or they can occur separately.
Hope this helps.
A word of caution. Do not use gPPI or PPI when the trial durations are set to 0. If you do, then you are testing the interaction of only the first 1/#time bins (default 1/16 or 1/8second). Anything that you are actually trying to capture in an interaction probably lasts more than 1/8 second. Thus you should model that duration of the interaction. The duration will have little effect on the shape of the HRF. The betas will change dramatically due to scaling of the HRF, but the statistics will be similar.
(2) An activated region is a region where the signal changes similarly across all trials of each trial type according to the HRF. A connected region is a region where the signal changes similarly across all trials of each trial type according to the neural activity in the seed region convolved with the HRF. As you can see, these are two fundamentally different signals. They can occur together or they can occur separately.
Hope this helps.
A word of caution. Do not use gPPI or PPI when the trial durations are set to 0. If you do, then you are testing the interaction of only the first 1/#time bins (default 1/16 or 1/8second). Anything that you are actually trying to capture in an interaction probably lasts more than 1/8 second. Thus you should model that duration of the interaction. The duration will have little effect on the shape of the HRF. The betas will change dramatically due to scaling of the HRF, but the statistics will be similar.