Hi, Joy-
Attached is an image of the initial positions/overlaps of the dsets (anat_epi_init_positions.png), and an image of the successfully aligned ones (from the QC HTML: align_epi_anat.png).
So, in this case, allowing the larger parameter space search of alignment values with "-giant_move" ended up leading to badness. I think it might be because both the anatomical and EPI are have pretty inhomogeneous brightness patterns (and in different ways). So, looking at your dsets, I figured that once there was a center-of-mass alignment, there wouldn't need to be a large amount of other rotation/shift/etc. done. So, I changed the option for align_epi_anat.py to this simpler one:
-align_opts_aea \
-cmass cmass \
... and that seemed to work (again, see attached alignment).
Note that I might also suggest adding:
-anat_has_skull yes \
... to your script (I do so followign the "-copy_anat ..." opt), to signify that the anatomical needs to be skullstripped; I think this is default behavior anyways, but I find it helpful to myself, because I usually skullstrip the anatomical *before* running afni_proc.py with @SSwarper, and then I remember to turn *off* skullstripping. (I generally like to *overspecify* options, just so I can adapt processing scripts more readily.)
One final note: I think we generally recommend not masking the regression step, so you can see the modeling results everywhere, even outside the brain, for QC and information purposes; we calculate masks at the single subject level, yes, but just use those to be combined at the group level for final results reporting. So, I would recommend not including this opt:
-mask_apply anat \
--pt