wengong-jin/refinegnn

refinegnn's Issues

What kind of input data does this take?

I'm not as familiar with antibody data. Is it via a molecular representation? Something else?

the CDR length n

Thank you very much for such inspiring work.
There is a small problem about CDR length n. The article mentioned in the pseudo code on page 6 that the length n of the CDR region was first predicted. However, I saw in the code that when facing section 4.2, RefineGNN used the real CDR length to generate the CDR-H3 sequence&structure for rabd.
Is there anything I missed？or is it reasonable.

Toward the biological meaning of generating CDR region with remaining sequence?

Hi, thanks for the great work!
I have a problem toward the biological meaning of the work. Since the CDR region is antigen-specific, how can we generate CDR sequence and structure with only remaining sequence give and without antigen-conditioned?
I hope you can help me with my confusion.

do we need to train a mAb/antigen pair to redesign the CDRs?

We have a few cases of mAb/antigen pairs where we know there is or isn't binding, but we don't know the exact binding mechanism (epitope/paratope).

If we want to increase binding or stability of a mAb/antigen pair with RefineGNN, starting with an Alphafold multimer pdb of the mAb Fv and the antigen (which may be wrong), how do we run RefineGNN to redesign the CDRs to increase the affinity/binding?
Do we need to train our mAb/antigen pairs to redesign the CDRs?

a questions about SARS-CoV-2 neutralization optimization

Thanks for your amazing research. Could you please share the hyper-parameters which can reproduce this experiment? I would appreciate it.