Examples and more benchmarking

[jadolfbr]

Hello! Really nice paper and work!
I was wondering if you had any examples of how to use your model? Also, it seems essentially that the model can do sequence design in the context of structure, with some structural variation. This could be really great for affinity maturation or homologous redesign. Have you compared the sequence recovery to ProteinMPNN? How does the recovery look on the canonical CDRs? Yes, H3 is important - but the other CDRs can matter quite a bit too, especially for certain epistles or peptides (in the case of L1). Does your model change the length of any of these CDRs, and do the canonical CDRs stay within the North clusters, or can they do cluster flips?

Cheers!
-Jared