Embedding Multichain proteins with ESM3 and ESMC

[Aurelien-Pelissier]

Hi,

Many proteins naturally exist in pairs, often referred to as alpha and beta chains, such as TCRs, antibodies, and MHCs. I’m curious about how ESM3/ESMC processes these cases.

• Does ESM3/ESMC have a special separator or handling mechanism for paired chains? I've seen the '|' separator in the sequence tokenizer.
• When using ESMProtein.from_pdb(pdbID), how does ESM handle multi-chain proteins?

Would appreciate any insights on this!

Thanks!

[SimonYYS]

I am also curious about whether some examples (for instance, 3_gfp_design.ipynb) could possibly include multimers, since I found there is "esm3-medium-multimer-2024-09" available, and dealing with multi-chain proteins are of much greater interest for researchers. Thanks!