Hi, Marina-

Are you using afni_proc.py (which is highly recommended for organizing your pre-processing)? During that processing, your regression model should contain all your baseline/drift regressors, motion time series, physiological regressors and bandpassing information, as well as your stimulus components (if doing task FMRI). That way, everything is done consistently, via a single model.

If doing a task study, I am not sure why bandpassing would be involved. Typically, one includes polynomial regressors to catch some of the slow/underlying drift of the signal (a type of regressor that some other softwares include via "high passing" above a very low frequency; those two methods should be essentially equivalent).

For resting state, it is common in the literature to bandpass, say, 0.01-0.1 Hz, or 0.001-0.1 Hz, or 0.01-0.08 Hz... something in those ranges. However, one main, reason people that people purport to do that is to try to filter out breathing+heartrate effects from the output. If you have physiological measures, that decreases the need for doing so. Additionally, groups have noted that a lot of useful signal exists still above 0.1Hz (see Gohel & Biswal, 2015, and Gohel et al., 2018, for example). Also, bandpassing over such a range reeaaaallly reduces the number of degrees of freedom remaining in the time series. So, I think you should ponder if you reeeeaaallly want to include that. Will it benefit your data, or will it just reduce the degrees of freedom for not much gain?

--pt