I am assuming that you are asking about the metadata provided with the RNA-Seq data downloads on https://portal.brain-map.org/atlases-and-data/rnaseq . Please let me know if you are referring to other gene expression matrices in the Allen Brain Atlases.
In the transcrip.tome HDF5 file for each of the mouse and human data sets, there are region and roi ids and labels. Unfortunately, the ids do not correspond to any structures in our reference atlas (provided by the “structure_id” field in the SVG path).
This is because the matching of a tissue dissection to a specific brain region can only be done on such a coarse scale. In fact, this is the very reason that we are currently pursuing a spatial transcriptomics strategy for future experiments - so that we can much more accurately map gene expression to a precise location in the brain.
The best mapping available is provided by the region and roi labels provided in the metadata. These correspond to brain regions, but not in a robust fashion suitable for making programmatic associations.