Dear Phil,

> someone replied to an older post of mine on this topic, and I'm not sure if my newer post is still showing as unanswered in the list, hence I'm forwarding it, if you don't mind (see post below).
> I think I can summarize what's confusing to me in a very short paragraph. 
> 
> The way I think of this is like this:
> 
> I have a timeseries from a seed region that responds HIGH during task A and LOW during task B.
> I take the dot product between this timeseries and a PSY task regressor that shows +1 during task A and -1 during task B. 
> Shouldn't the dot product (the PPI interaction regressor) thus still show HIGH when the timeseries shows HIGH? We're just multiplying by +1.
> 
> What I am confused about now is this: why is this PPI regressor then pulling out brain regions that show LOW during task A?

are you sure that your experiment is an ideal candidate for PPI? Very often PPI experiments are performed by having a stimulation/task of some sort and then a "psychological state manipulation" that alters ones "state" when performing that task.

I believe the classical example is Christian Buchels study where he had moving colored dots on the screen and once in a while the dots would change (very subtly) either their speed or their color. The stimulus (the changes) was identical throughout the experiment and the psychological state change was blocks where the subjects were instructed to either focus on detecting speed changes or focus on detecting color changes (apologies if I don't remember it completely correctly).

We have here an ideal situation where the stimulus (the changes) elicits a response in V1 and then we can use PPI to see if there is a difference in correlation between V1 and e.g. V5 when focusing on motion compared to when focusing on color.

In a multifactorial setting like this is makes sense to one factor (focus) to modulate the activity in a seed region elicited by the other factor.

If you only have a single factor (A vs B) that elicits activity in your seed region and then uses the same factor to modulate that, it becomes a bit strange. It sounds like your seed region is "high" during A and "low" during B. If you know take that and modulate that by the same task it sounds like you get something that is high all the time and which can hardly be very interesting to correlate anything against.

Do you understand my argument about the two factors?

Jesper

 
> 
> I'm hoping it's not a bug either in my scripts or in the software
> 
> Many thanks,
> 
> Phil
> 
> 
> ----- Forwarded Message -----
> From: Phil Yoss <[log in to unmask]>
> To: FSL - FMRIB's Software Library <[log in to unmask]>
> Sent: Monday, June 27, 2011 6:58 PM
> Subject: Re: [FSL] PPI results question
> 
> Dear Jesper,
> 
> thanks a lot for your reply. If I can try just one more time (I'm still not sure I'm quite clear on this):
> 
> my seed region is behaving such that it is highly active during task A. I.e. the PHYS regressor shows high activity during task A.
> 
> So I want to find out what other brain regions correlate with my seed region, but especially during task A, not task B. So I have a PSY regressor A-B. I take the dot product between the two and obtain my PPI regressor (in FSL, using the software as instructed).
> 
> Now: I identify several regions correlating with this PPI regressor, which are highly significant. If indeed their correlation with the seed region goes up during task A, and it is a positive correlation, shouldn't these PPI regions behave similar to the seed region during task A? I.e. if my seed region is highly active during task A, and I have a task-specific correlation, shouldn't the PPI region also be positively activated during task A?
> 
> Instead, what I find when I examine the PPI ROIs and extract the PEs/percent signal change for the PPI regions from the full GLM, is that several of those regions are actually *deactivated* during task A (and highly positive during task B).
> 
> (Perhaps what's confusing is this: I first indentify the PPI regions. Then I select those voxels, for each PPI ROI, and check in the full GLM how those voxels behave during each condition -- by looking at the PEs and calculating % signal change).
> 
> I thought if it's a task-specific correlation with my seed region, then during task A, both the seed region and the PPI region should be similar to each other -- i.e. both should be positively activated during task A. (Otherwise how can their correlation increase during task A, if seed region X and PPI region Y are not becoming more similar in their activation?)
> 
> Am I wrong? Or is it simply the case that the PPI is showing me both positive correlations with my seed region during task A, as well as negative correlations with my seed region during task A -- i.e.:
> 
> seed region is +ve during task A, but PPI region is -ve during task A?
> seed region is less +ve during task B than A, but PPI region is more +ve during task B than A?
> 
> Sorry if I'm missing something, I am really trying to understand this. 
> 
> Thanks a lot,
> 
> Phil
> 
> From: Jesper Andersson <[log in to unmask]>
> To: [log in to unmask]
> Sent: Monday, June 27, 2011 10:49 AM
> Subject: Re: [FSL] PPI results question
> 
> Dear Phil,
> 
> I struggle a little to understand precisely your description below of what you see in your data.
> 
> In general in a PPI you would not expect to find something that simply correlates with your seed region, nor which correlates with your task (psychological  processing). What you are looking for is an area where the correlation with the seed is different during one condition compared to during another. It is often not obvious from just looking at the time-series and the full model fit what is happening since the full model will include also the "raw" seed region and the contribution from the PPI regressor will often just be to modulate this a little bit so that the correlation is a little stronger during one of the conditions than during the other.
> 
> The PPI regressor itself is constructed so that the activity from the seed region is sign reversed during one of the conditions compared to during the other. However, as outlined above, this is very rarely what one sees in the actual data (the sign reversal). Instead the PPI regressor is just modulating a positive (or negative) correlation.
> 
> I hope this was more helpful than confusing.
> 
> Good Luck Jesper
> 
> 
> On 27 Jun 2011, at 08:43, Phil Yoss wrote:
> 
>> Thanks, Donald. I still don't quite understand this though. Sure, the seed region causes up and down fluctuations -- but relative to what? Relative to baseline. So in seed region X, when task A is happening, there is an upward fluctuation, and when task B is happening, there is a down-ward fluctuation. Shouldn't I be seeing the same pattern in the regions the PPI pulls out, precisely the regions that supposedly correlate with the seed region?
>> 
>> Having looked at my data more carefully, I think I am now essentially seeing just two patterns in the regions produced by the PPI:
>> 
>> 1) task A > task B
>> task A = +ve (above baseline), task B less +ve (but still above baseline); same as task A  =+ve, task B = 0 or -ve;
>> [this is consistent with the PPI, since task A is more active than task B]
>> 
>> and
>> 
>> 2) task A<task B:
>> task A = +ve (above baseline), task B even more +ve; same as task A = 0, task B = +ve; same as task A = -ve, task B = +ve.
>> (in all of these task B is basically more active than task A)
>> 
>> So I'm now wondering about a new issue: does the PPI correlation pull out both *positive* and *negative* correlations with the seed region? But in this case, how come I see both positive and negative correlations for a positive threshold, e.g. z=3.1? Wouldn't I have to set z = -3.1 to see the anticorrelated areas?
>> 
>> Many thanks,
>> 
>> Phil
>> 
>> 
>> From: "MCLAREN, Donald" <[log in to unmask]>
>> To: [log in to unmask]
>> Sent: Sunday, June 26, 2011 11:33 PM
>> Subject: Re: [FSL] PPI results question
>> 
>> PPI is fundamentally different the task activity.
>> 
>> PPI explains the covariation in signal that is in addition to the
>> task. One way to think about PPI is to think of the BOLD response as
>> smooth and then think of another region cause fluctuations up and down
>> from that smooth response and that these fluctuations are the result
>> of the activity in the seed region.
>> 
>> Thus, the task activity is separate from the PPI activity.
>> 
>> Best Regards, Donald McLaren
>> =================
>> D.G. McLaren, Ph.D.
>> Postdoctoral Research Fellow, GRECC, Bedford VA
>> Research Fellow, Department of Neurology, Massachusetts General Hospital and
>> Harvard Medical School
>> Office: (773) 406-2464
>> =====================
>> This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
>> HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
>> intended only for the use of the individual or entity named above. If the
>> reader of the e-mail is not the intended recipient or the employee or agent
>> responsible for delivering it to the intended recipient, you are hereby
>> notified that you are in possession of confidential and privileged
>> information. Any unauthorized use, disclosure, copying or the taking of any
>> action in reliance on the contents of this information is strictly
>> prohibited and may be unlawful. If you have received this e-mail
>> unintentionally, please immediately notify the sender via telephone at (773)
>> 406-2464 or email.
>> 
>> 
>> 
>> 
>> On Sun, Jun 26, 2011 at 1:48 PM, Phil Yoss <[log in to unmask]> wrote:
>> > Hello all,
>> > is it possible to obtain highly significant PPI results (i.e. areas that
>> > significantly correlate with a psycho-physiological interaction regressor)
>> > which show a very different response profile than the seed region going into
>> > the PPI regressor? I.e. if my PPI regressor shows high activity on task A
>> > and low activity on task B, why is it that I am getting regions correlating
>> > with this PPI regressor, but which show e.g. negative activity on task A and
>> > B, or negative activity on task A but positive on task B, or positive on
>> > both task A and B? (I am plotting the % signal change for the areas pulled
>> > out by the PPI).
>> > I don't understand why the BOLD response profile varies so much in areas
>> > correlating with the PPI regressor, and why the response profile (% signal
>> > change by condition) doesn't look like the PPI regressor (i.e. high on task
>> > A, low on task B). The areas in question survive thresholding at p < 0.005
>> > or lower, corrected. Am I doing something wrong?
>> > Thanks much,
>> > Phil
>> 
>> 
> 
> 
> 
> 
>