Hi Dr. McLaren,

I have run first- and second-level analysis using PPI and found some interesting results (thank you for the toolbox). Just to get a better idea of what I'm looking at, I would like to visualize the results for each subject and for the 2nd-level analysis (I have a group comparison analysis done) using scatter plots. I have seen instructions to create scatter plots in Gitelman's PPI chapter in the SPM manual but I suspect that it doesn't work with PPI. Could you please point me to the right direction with regard to creating such plots?

Your help is much appreciated.

Regards,

Jeff

There isn't an easy out of the box way to plot the interaction for event-related designs as the BOLD response for neighboring events overlaps. The method in the PPI chapter, if I remember correctly, is to plot the 2 PPI terms against each other. This is only a rough approximation and the slope won't be the same as the actual data.

If you want to actually make a plot reflective of the actual analysis at the first level, you would need to do the following for each condition:
(1) Regress all the columns of X except the PPI column of interest from the PPI column of interest. This gives you the residual of the PPI column.
(2) Regress all the columns of X except the PPI column of interest from the BOLD signal in the target region. This gives you the residual of the BOLD signal not accounting for the PPI column.
(3) Plot the results of (1) against (2).
(4) Repeat steps (1) to (3) for each condition.

The slope of the line will be the same as it is at the group level for that subject, as you've regressed out the mean connectivity, so I'm not sure it will be much help.

The betas for each condition are the change in connectivity relative to baseline.

Hi Dr. McLaren,

Thank you for your answer. Just so I know I understand you correctly:

- Step 1 refers to the extraction of the PPI component from the seed region, step 2 refers to the extraction of BOLD component from the 2nd region that shows connectivity with the seed region. Would this be correct?
- The betas come from the SPM.Vbeta. When I look at the description field, there are conditions (e.g, low_load, high_load) and PPI_conditions (e.g., PPI_low_load, PPI_high_load). I'm assuming the former is the BOLD component and the latter is the interaction component which are extracted in the step 1 and 2 above. Could you please confirm this?

Regards,

Jeff

Yes. I believe this is correct.

You have your gPPI model with a number of columns in the design matrix, step 1 means to regress all other columns out of the PPI column. In other words, you create a new design matrix without the PPI column, then you use the PPI column as the DV, to produce the residuals.

I=column of interest
DM is the design matrix.
PPIres=DM(:,I)-pinv(DM(:,setdiff(1:size(DM,2),I))*DM(:,I);

Step 2 replace DM(:,I) with the BOLD signal from the target region.

Yres=Y-pinv(DM(:,setdiff(1:size(DM,2),I))*Y;


Plot PPIres against Yres.

======
The answer to your second question is No. While, VBeta contains the beta estimates of the full model; Step 1 and Step are computed separately from the original estimates. See above comment.