Oh, I left one thing out of that --- even with the underscore issue aside, the table's features are created from OTU clustering using sh_refs_qiime_ver7_dynamic_s_01.12.2017.fasta, which uses a slightly different ID scheme than the sh_refs_qiime_ver7_dynamic_01.12.2017.fasta file in the other variant. So even if the underscores were quoted, the IDs just aren't quite the same.
import skbio
sh_refs_qiime_ver7_dynamic = skbio.io.registry.read('sh_refs_qiime_ver7_dynamic_01.12.2017.fasta', format='fasta')
sh_refs_qiime_ver7_dynamic_s = skbio.io.registry.read('sh_refs_qiime_ver7_dynamic_s_01.12.2017.fasta', format='fasta')
sh_refs_qiime_ver7_dynamic_ids = {s.metadata['id'] for s in sh_refs_qiime_ver7_dynamic}
sh_refs_qiime_ver7_dynamic_s_ids = {s.metadata['id'] for s in sh_refs_qiime_ver7_dynamic_s}
print(len(sh_refs_qiime_ver7_dynamic_ids))
print(len(sh_refs_qiime_ver7_dynamic_s_ids))
print(len(sh_refs_qiime_ver7_dynamic_ids.intersection(sh_refs_qiime_ver7_dynamic_s_ids)))
The last three lines return:
30696
58049
29818
So there is some overlap between the two variants of that DB, but not full overlap. This will cause problems in a variety of places if you mix and match between the two.