help
help > RE: including a parametric modulator
Sep 12, 2018 09:09 PM | rcooper1
RE: including a parametric modulator
Hi Alfonso,
I am trying to run a gPPI analysis in CONN (ROI-to-ROI) using parametric modulators and I have followed the instructions you provided below about entering condition-specific temporal modulators. Everything seems to be running fine but I would like to check the gPPI first level design matrices to make sure that my regressors look as expected, but I can't seem to fine this information anywhere?
The parametric modulators I have are continuous, trial-specific measures of memory performance. I extracted them from my first level SPM.mat files, SPM.xX.X, which contains the mean-centered parametric modulators to be used in my standard univariate first-level models. I specifically want to check that:
1) my parametric modulators (added as first-level covariates in CONN) have been correctly used to weight by my 'conditions' (I just added these as blocks with 0 onset and inf duration, as suggested below), and ...
2) my covariates are being including as regressors in the gPPI analysis (to ensure that both my psychological and physiological variables are controlled for when calculating each ppi).
Any help would be much appreciated!
Best,
Rose
Originally posted by Alfonso Nieto-Castanon:
I am trying to run a gPPI analysis in CONN (ROI-to-ROI) using parametric modulators and I have followed the instructions you provided below about entering condition-specific temporal modulators. Everything seems to be running fine but I would like to check the gPPI first level design matrices to make sure that my regressors look as expected, but I can't seem to fine this information anywhere?
The parametric modulators I have are continuous, trial-specific measures of memory performance. I extracted them from my first level SPM.mat files, SPM.xX.X, which contains the mean-centered parametric modulators to be used in my standard univariate first-level models. I specifically want to check that:
1) my parametric modulators (added as first-level covariates in CONN) have been correctly used to weight by my 'conditions' (I just added these as blocks with 0 onset and inf duration, as suggested below), and ...
2) my covariates are being including as regressors in the gPPI analysis (to ensure that both my psychological and physiological variables are controlled for when calculating each ppi).
Any help would be much appreciated!
Best,
Rose
Originally posted by Alfonso Nieto-Castanon:
Hi Erik,
If you already have those parametric-modulation timeseries as first-level covariates "param_conf_A" and "param_conf_B" (and I guess you probably also have the original conditions onsets/durations entered as conditions "A" and "B"), what I would suggest is the following:
1) create two conditions associated with these covariates. E.g. create a new condition "conf_A", enter 0/inf in the onset/duration fields in all sessions where the condition is present, then select in the 'optional fields. task modulation factor' field the option "condition blocks * covariate param_conf_A". The same for the new condition "conf_B" (now select task modulation factor param_conf_B)
2) create a new first-level analysis with the 'analysis type' field set to gPPI, and select when prompted the conditions "A", "B", "conf_A", and "conf_B"
3) in the second-level results window select all four conditions and enter the contrasts:
[0 0 1 0] to look at the association between connectivity and confidence scores in condition A
[0 0 0 1] to look at the association between connectivity and confidence scores in condition B
[0 0 1 -1] to look at the difference in association between connectivity and confidence scores between the two conditions
[1 0 0 0] to look at the relative connectivity during condition A (average connectivity compared to baseline at the zero-level of the confidence scores covariate)
[0 1 0 0] to look at the relative connectivity during condition B (average connectivity compared to baseline at the zero-level of the confidence scores covariate)
[1 -1 0 0] to look at the average connectivity differences between the two conditions (both estimated at the zero-level of the confidence scores covariate)
Last, regarding your question (1), yes, in almost all cases you want to include the "effect of *" covariates as confounding effects during Denoising (this controls for the direct association between the BOLD signal and the covariate, it does not remove or control for the association between the covariate and connectivity values which is what one typically cares about in connectivity analyses). In this case, since you are going to be doing PPI analyses (which explicitly control for the same regressors -the "main psychological term"- in the PPI equation) it should not make a difference whether you enter them or not in the Denoising step (but I would still recommend doing so just for consistency across analyses).
Hope this helps
Alfonso
Originally posted by erik wing:
If you already have those parametric-modulation timeseries as first-level covariates "param_conf_A" and "param_conf_B" (and I guess you probably also have the original conditions onsets/durations entered as conditions "A" and "B"), what I would suggest is the following:
1) create two conditions associated with these covariates. E.g. create a new condition "conf_A", enter 0/inf in the onset/duration fields in all sessions where the condition is present, then select in the 'optional fields. task modulation factor' field the option "condition blocks * covariate param_conf_A". The same for the new condition "conf_B" (now select task modulation factor param_conf_B)
2) create a new first-level analysis with the 'analysis type' field set to gPPI, and select when prompted the conditions "A", "B", "conf_A", and "conf_B"
3) in the second-level results window select all four conditions and enter the contrasts:
[0 0 1 0] to look at the association between connectivity and confidence scores in condition A
[0 0 0 1] to look at the association between connectivity and confidence scores in condition B
[0 0 1 -1] to look at the difference in association between connectivity and confidence scores between the two conditions
[1 0 0 0] to look at the relative connectivity during condition A (average connectivity compared to baseline at the zero-level of the confidence scores covariate)
[0 1 0 0] to look at the relative connectivity during condition B (average connectivity compared to baseline at the zero-level of the confidence scores covariate)
[1 -1 0 0] to look at the average connectivity differences between the two conditions (both estimated at the zero-level of the confidence scores covariate)
Last, regarding your question (1), yes, in almost all cases you want to include the "effect of *" covariates as confounding effects during Denoising (this controls for the direct association between the BOLD signal and the covariate, it does not remove or control for the association between the covariate and connectivity values which is what one typically cares about in connectivity analyses). In this case, since you are going to be doing PPI analyses (which explicitly control for the same regressors -the "main psychological term"- in the PPI equation) it should not make a difference whether you enter them or not in the Denoising step (but I would still recommend doing so just for consistency across analyses).
Hope this helps
Alfonso
Originally posted by erik wing:
Hi Alfonso, many thanks for this info. I had a
follow up question about the parametric modulator analysis. In my
spm.mat file I have two event-related conditions A and B, each of
which has a trialwise parametric modulator that codes continuous
ratings of confidence. For each condition I extracted the estimated
parametric regressor column from the SPM.xX.X and entered it as a
first-level covariate, resulting in two additional covariates
(param_conf_A, param_conf_B) along with the standard 'SPM
covariates'. If the goal is to look at connectivity differences
between the confidence-modulated A and B conditions (and if the
setup sounds OK up to this point), I was wondering 1) does it make
sense to omit the parametric modulator covariates from the
Confounds list in the denoising step, or should they be kept in? 2)
is it necessary to run the first-level analysis twice with a
different modulator selected under 'other temporal-modulation
effects' for each. It seems like only one interaction factor can be
selected. I was also trying to determine which option under
first-level 'Analysis options' makes most sense. Thanks,
erik
erik
Threaded View
| Title | Author | Date |
|---|---|---|
| Richard Morris | Jul 20, 2012 | |
| Alfonso Nieto-Castanon | Jul 20, 2012 | |
| Ewa Miendlarzewska | Oct 12, 2015 | |
| Alfonso Nieto-Castanon | Oct 14, 2015 | |
| Katherine Swett | Apr 6, 2020 | |
| Tina Tasia | Jun 9, 2021 | |
| Shady El Damaty | Nov 11, 2016 | |
| Alfonso Nieto-Castanon | Nov 12, 2016 | |
| erik wing | Dec 13, 2016 | |
| Alfonso Nieto-Castanon | Dec 14, 2016 | |
| rcooper1 | Sep 12, 2018 | |
| erik wing | Dec 20, 2016 | |
| Michael Jacob | Oct 3, 2017 | |