Technically this isn't a product of Multimer, but rather a roundabout way using AF2 to predict homomers. Given that most unique oligomers in pdb are symmetrical homomers, with this caveat it sorta works. Although I threw a couple of known homodimers against it that ain't got templates and obv got 0.

Did they compare AF and multimer in the paper for a few cases? Would be interesting indeed