Hi Dane,
I've been out/busy for a while, sorry...
A am not sure in exactly what way you are using NaN, but
AFNI does not use that as a special case. It is likely
you should not use it at all.
Regressing out everything but one condition at a time is
a little shaky, particularly for a fast event design,
since regressors are much less likely to be orthogonal.
Consider instead applying the full regression, and then
adding the signal of interest to the errts. That could
be done by adding the -cbucket option to 3dDeconvolve
(see -regress_make_cbucket) and running 3dSynthesize
afterwards. For classes with single regressors, it is
also easy enough to just use 3dcalc (beta volume times
regressor of interest plus noise).
As a side note, reviewers may not be very accepting of
this method, as most of the correlations would possibly
be driven by the main response to the regressor of
interest.
For bandpassing, look up -regress_bandpass in the
afni_proc.py -help output (or -h_view). That option
is used in most of the resting state example, such as
example #9.
If you are just computing correlations, then scaling is
not relevant. However it does not hurt, and since you
are still doing a task regression, it might be good to
stick with it so the intermediate results make more sense.
Using 3dTproject on the regression matrix gains you
nothing but a lot of saved time (it is much faster).
Again, just look at what afni_proc.py does for the case
of bandpassing with no tasks.
Regarding your afni_proc.py command:
0. You have not specified any blocks, meaning this will
do time shifting (requires an accurate TR), volume
registration, blurring scaling and regression.
1. Pass it all sessions at once. If you do not, the EPI
will not be aligned across runs. The runs do not need
to be of the same length.
2. There is no -force_TR option. Why do your datasets
not already have that TR set? You can use 3drefit to
add it or stick with passing -force_TR to 3dDeconvolve
via -regress_opts_3dD.
3. Since you are using TENT functions, I suggest going the
-regress_make_cbucket and 3dSynthesize route.
4. A -scale_max of 100 makes no sense, since 100 is the
default mean. Stick with 200 or omit that option.
- rick