See in line responses below.

On Mon, Dec 10, 2012 at 10:30 AM, Annchen Knodt <[log in to unmask]> wrote:
> Dr. McClaren,
>
> Thanks again for providing help with PPPI analyses. We used your toolbox and
> have some interesting results, but we have questions about how to interpret
> and probe these effects.
>
> What we've done is enter the conditions (specified earlier in the thread)
> into the design matrix for individual subjects and produce PPPI images. We
> then enter the PPPI images into a one sample T test, to see if the beta
> values for the PPPI images across the group are different than zero, and get
> significant clusters.
>
> From there, how do we interpret the interaction? If we start with thinking
> about the simpler psycho-physiological interaction (PPI), we know that the
> values of the parameters describe the difference in slopes (for example:
> that the correlation between the seed region and identified voxels is more
> positive during condition A than during condition B); however, that
> difference in slopes could mean different things for different subjects (One
> subject could have positive patterns of connectivity during both conditions,
> with one condition that is significantly more positive than the other,
> however another subject might have a pattern of positive connectivity during
> one condition, and a non-positive correlation during another condition).
> How would you suggest probing such PPI interactions to make inferences about
> group effects?

This is an excellent question. In traditional PPI, one would only get
a difference in the slope between two conditions and simply stop at
that level. With the advent of gPPI, one can probe the differences
between conditions relative to the implicit baseline. For example, you
could have A>B and both A and B>baseline or A and B both less than
baseline. Then, you end up with the question, how do you interpret
A>baseline. All this means is that the connectivity slope is more
positive. It can't tell you if the connectivity overall is
positive/negative. Future work will probably look at using the
implicit baseline term to determine the overall connectivity of each
condition.

>
> Further, would you have any recommendations about how to probe interactions
> for PPPI analyses?

Along with probing the differences, also look at each condition versus
baseline. This combination should be sufficient to better understand
how the task changes connectivity.

Hope this helps.

>
> Any help would be appreciated. Thanks again.
>
> Annchen
>
>
>
> On Nov 12, 2012, at 5:43 PM, MCLAREN, Donald wrote:
>
> The issue of collinearity is most prominent in event-related designs
> or in three-way psychophysiophysiological interactions when the neural
> activity in the seed regions is almost identical.
>
> Here is an example. If I have 1 trial in one run of an event-related
> design (event duration 0), then the PPI regressor will be identical to
> the psychological regressor. If there are two events, and the neural
> activity for both events is the same, then the PPI and psychological
> regressors will be identical again. As you add more trials (or switch
> to a block design), the probability that the neural activity is
> constant during anytime there is an event drops dramatically. Then,
> the PPI and psychological vectors won't be collinear. If the seed
> regions are too similar, then the PPI vectors for seed*task
> interactions will be collinear as well. My estimations is that you
> need at least 5-10 events per run to get away from the collinearity if
> your event duration is 0. The more trials the better.
>
> Hope this helps.
>
> It is unrelated to the warning message that you see. That warning
> message is related to the deconvoution step.
>
> Best Regards, Donald McLaren
> =================
> D.G. McLaren, Ph.D.
> Research Fellow, Department of Neurology, Massachusetts General Hospital and
> Harvard Medical School
> Postdoctoral Research Fellow, GRECC, Bedford VA
> Website: http://www.martinos.org/~mclaren
> 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 Mon, Nov 12, 2012 at 2:01 PM, Annchen Knodt <[log in to unmask]> wrote:
>
> Hi Donald,
>
>
> We got the PPPI toolbox up and running and it does look like it will provide
> us with the modeling capability we need.
>
>
> I am wondering if you might elaborate on your concern about collinearity in
> the design, though.  We're getting the following warning in MATLAB and am
> wondering whether such a collinearity might be responsible:
>
>
> Warning: Matrix is close to singular or badly scaled.
>
>         Results may be inaccurate. RCOND = 1.266417e-020.
>
> In spm_PEB>spm_inv at 371
>
>  In spm_PEB at 244
>
>  In PPPI at 633
>
>
> Our design matrix has 7 columns as you specified below (plus one for the
> constant) - we only have one task which has a block design consisting of 5
> control blocks interleaved with 4 experimental blocks.  Where exactly would
> problems with collinearity arise?  Let me know if you need more info about
> the task.
>
>
> Thanks so much!
>
>
> Annchen
>
>
>
>
> On Oct 23, 2012, at 1:29 PM, MCLAREN, Donald wrote:
>
>
> Please see inline responses below.
>
>
> On Tue, Oct 16, 2012 at 2:22 PM, Annchen Knodt <[log in to unmask]> wrote:
>
> Hello All,
>
>
> I would like to model a physio-physiological interaction in SPM 8
>
> (interaction of time series from 2 separate ROIS), however I would like to
>
> determine if that interaction is dependent on the psychological condition
>
> determined by the experimental design. Stated differently, I want to test if
>
> the physio-physiological interaction during condition A is significantly
>
> different than the same interaction in condition B. My first concern is to
>
> make sure I am modeling this question properly. The second issue regards
>
> implementation.
>
>
> 1)     If I understand the SPM8 manual properly, in order to test this
>
> question I should enter the following predictors into the design matrix:  1)
>
> a regressor representing the convolved time series from ROI 1; 2) a
>
> regressor representing the convolved time series from ROI 2; 3) a condition
>
> representing the experimental design; 4) a regressor representing the
>
> convolved time series from ROI 1 multiplied by the experimental design; 5) a
>
> regressor representing the convolved time series from ROI 2 multiplied by
>
> the experimental design; 6) a regressor representing the convolved time
>
> series from ROI 1 multiplied by the convolved time series from ROI 2; 7) a
>
> regressor representing the convolved time series from ROI 1 multiplied by
>
> the convolved time series from ROI 2 and the experimental design (the three
>
> way interaction term).
>
>
> Is this correct?
>
>
> Almost. You want the following regressors:
>
> (1) timeseries from ROI1 (BOLD signal extracted from the ROI)
>
> (2) timeseries from ROI2 (BOLD signal extracted from the ROI)
>
> (3) Your task design matrix (one column for each task)
>
> (4) the convolution of (the deconvolved timeseries in ROI1 * task) -->
>
> 1 column for each task
>
> (5) the convolution of (the deconvolved timeseries in ROI2 * task) -->
>
> 1 column for each task
>
> (6) the convolution of (the deconvolved timeseries in ROI1 * the
>
> deconvolved timeseries in ROI2)
>
> (7) the convolution of (the deconvolved timeseries in ROI1 * the
>
> deconvolved timeseries in ROI2 * task) --> one column for each task
>
>
> This can be done automatically with the gPPI toolbox; however, nothing
>
> has been published on the three-way interaction as far as I know. I
>
> also think you will need a number of trials in your task to get the
>
> models to estimate the effects accurately to avoid collinearity in
>
> your design.
>
>
>
>
> 2)     Finally, does anyone know if this can be implemented in SPM8 using
>
> the Batch function? Additionally any other advice about implementation would
>
> be appreciated.
>
>
> Use the gPPI toolbox available through NITRC (www.nitrc.org/projects/gppi)
>
>
>
> Thanks in advance for any help.
>
>
>
> Annchen Knodt
>
>
>
> ~~~~~~~~~~~~
>
> Annchen Knodt, M.S.
>
> Research Associate
>
> Laboratory of Neurogenetics
>
> 919.684.1039
>
> Duke University
>
>
>
>
>
>
>
>