Thanks for your reply, Rick.
A couple of things not clear to me.

Regardless, the goal is to find a fit to the interaction
term, perhaps while ruling out a fit to just the seed.
Even if the seed happened to be a combination of task
regressors, we would not care (multicolliearity aside),
because those terms are all considered "noise" in the
model. Only the interaction result matters.

- when I did the CDCor analysis, and overlayed the betas for the ROI, the map looks weird in the sense that
some measurement noise contaminating voxels aroudn the ROI all along the brain in AP and DV but not in LR directions.
The interaction betas make sense. SHould the betas for ROI not reflect correlation of the ROI time course with all other voxels above and beyond the task regressors?


Along these lines, it is generally a nice idea to start
with the residual time series when extracting the seed
to begin with. There is no reason to leave the seed
contaminated with any signals that one might consider as
noise from the view of the new model.

So, As I understand First do the GLM with just the taks regressors and regressors of no interest.
get the ROI time series using the residual time series from ROI voxels from the above model.
Use the ROI residual time series to get interaction terms for task regressors.
Use the above ROI residual time series and the interaction terms as regressors to fit the residuals from the model with task regressors.
Is that a nice way to do it?

Amar