Sorry, there is no mechanism for that, as far as I know.
I would make a separate regressor class for all trials
with missing modulators, and not modulate those events
at all (i.e. only model the mean response). So one class
would have all modulators, the other would have none.
- rick