Hi, Joy-
The problem with chopping up the alignment to final/standard space into multiple steps, each creating intermediate datasets, is that smoothing occurs at each regridding process. So, you introduce unnecessary smoothing at each point. That is why we put all the stages into afni_proc.py, and the final transformation from EPI to final space is performed at once (all intermediate alignments are concatenated and applied in one go, rather than trickling through separately).
I am not sure why a GLM would need to be performed in native space?
For going to standard space, using nonlinear alignment will give *much* better correspondence across a group analysis than using @auto_tlrc (which is linear affine, only). Using @SSwarper would be a better way to go.
--pt