Thanks a lot Mike for this, we got the meaning of Donald's comment on that point now.
From this, how is it possible to disentangle these eventualities? More precisely, if one has a significant positive estimate for the interaction term for the contrast, say, {A - B}, in a cluster C. How to quantify the individual "coupling" between the seed region and cluster C separately for A & B?
(again, this seems to be done across various studies, e.g. Manelis et al., 2013, Hippocampus)
Best
Pierre-Yves & Quentin
________________________________________
De : Angstadt, Mike [[log in to unmask]]
Envoyé : mardi 3 avril 2018 17:52
À : [log in to unmask]; JONIN Pierre yves
Objet : Re: [SPM] PPI analysis and connectivity quantification
I think Donald's comment about values being difficult to determine is with respect to the specific conditions (A and B). Of course with PPI you can determine whether A>B or B>A, but this can be due to a number of situations (the same as in regular activation contrasts). B>A can mean: A positive, B more positive; or A flat, B positive; or A negative, B positive or flat; or A very negative, B less negative; etc. All of these situations are B>A, but just knowing B>A doesn't tell you anything about the directionality of A and B individually.
-Mike
________________________________________
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of JONIN Pierre yves <[log in to unmask]>
Sent: Tuesday, April 3, 2018 10:21:37 AM
To: [log in to unmask]
Subject: Re: [SPM] PPI analysis and connectivity quantification
Hi Donald & thanks a lot for your answer,
We'll go for a gPPI pipeline then, however we are unsure about the connectivity quantification. What we are looking for when refering to "strength" is actually a way to quantify the connectivity that allows further stats, like comparing connectivity across groups or conditions, or correlating connectivity with e.g. behavioural variables.
If we got you right, it seems that the parameter estimate scaling the Psycho-Physiological Interaction term associated with a given cluster can indeed be used to quantify the "coupling" between the seed region and a given cluster. This parameter represents a difference between conditions or between condition X and a null event, or it may also represents an interaction contrast (i.e. this parameter actually reflects whatever contrast was used to define the psychological term).
Since we do obtain (with sPPI at least, we have not yet run gPPI) negative AND positive results, we do not understand your comment about whether absolute values are positive or negative is hard to determine? If the psychological term included conditions A & B, with a contrast defined as {A > B}, then positive PPI outcome suggests a greater connectivity for A than for B (i.e. postive Beta for the interaction), right? Could you please clarify this, and the further comment about considering the physiological regressor to determine the sign of these absolute values?
We also have two follow-up questions:
- do you recommend to include null events when building the psychological regressor for the PPI?
- to further understand the quantification issue, we have read many studies referring to PPI, and often the results suggest to us that we clearly miss something. For example, Manelis et al. (2013, Hippocampus<https://onlinelibrary.wiley.com/doi/abs/10.1002/hipo.22053>), provide in their Figure 5 (p. 61) data suggesting that some quantification of the connectivity can be obtained separately for each condition of a contrast (see the Y-axis). The contrast used to build the psychological term is "{Presentation 2 - Presentation 1}", and the figure seems to provide separate connectivity estimates for each presentation....how is that possible, given that only one parameter estimates result from a typical PPI model derived from a simple {A - B} contrast ?
Sorry for such a long email, we really find it hard to get clear answers elsewhere....
Thanks a lot in advance,
Pierre-Yves & Quentin
________________________________
De : MCLAREN, Donald [[log in to unmask]]
Envoyé : samedi 31 mars 2018 00:04
À : JONIN Pierre yves
Cc : SPM
Objet : Re: [SPM] PPI analysis and connectivity quantification
Hi Pierre-Yves & Quentin,
Generalized PPI. This is the recommended approach anytime you have more than 2 conditions including null events as one of the 2 conditions.
It's the amplitude of the connectivity as its a scalar of the regressor. Whether this is "strength" or not depends on what you mean by strength. Additionally, its important to remember that the estimates represent a difference between 2 conditions or the condition and baseline if using gPPI. Whether the absolute values are positive or negative is harder to determine from the contrasts, but might be possible by considering the physiological regressor in the contrasts.
Future research is aimed at conducting more simulations of how the gPPI can extract the absolute connectivity rather than just relative connectivity at the moment.
Best Regards,
Donald McLaren, PhD
On Thu, Mar 29, 2018 at 11:25 AM, JONIN Pierre yves <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Dear SPM experts,
We have run an event-related fMRI experiment and analyzed data by setting up a 2 (type of stimuli) x2 (repetition) factorial design. Based on the clusters derived from our main effects and effect of interaction, we aimed at testing hypothesis regarding the functional connectivity from these data-driven seeds. That is, we have investigated the functional connectivity of seed regions derived from simple effects (i.e. effect of the type of stimuli, or effect of repetition) and the connectivity of seed regions derived from the interaction contrast.
Our questions are as follows:
- in that context, is standard PPI or generalized PPI best recommended?
- how can we estimate the "strength" of the functional connectivity? Our feeling is that the parameter estimate of the interaction term of the PPI analysis inside a resulting cluster can provide such an estimate of connectivity strength (e.g. see that study<http://www.jneurosci.org/content/32/29/10012>), but we are unsure about this interpretation.
Thanks in advance for your help,
All the best
Pierre-Yves & Quentin
**********************************************************
Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues
|
