Commercial License + combined mutations

[jadolfbr]

Hello,

The CC4.0 is a very restrictive licence for a project. I was wondering if you are sticking with that license (even though most of the code is a wrapper to using ESM-1b/v) or if you have any simple commercial licenses available through your tech transfer office.

I was also wondering why the subsequent mutations in your paper were manually put together. It seems like it would be possible to do a combinatorial scoring and order best-scoring hits from the manifold. Did I miss something there?

Thanks,
-Jared

[brianhie]

Thanks for reaching out @jadolfbr, and apologies for the late response!

I'll keep you updated about the license, and I'll certainly look into alternatives.

For the round 2 combinations, in the MEDI (UCA) trajectories we did just test all possible combinations. For the mAb114 and SARS-CoV-2 trajectories, combinations were chosen to get better diversity of edit distances than would be suggested by just, e.g., computing pseudolikelihoods, which tends to prefer just adding all of the mutations together, though this is indeed possible to do. Manually combining single-site mutations based on a first round is common in traditional engineering campaigns and was a natural choice to explore a more diverse subset of the combinatorial space in a single round given our very low-throughput pipeline. With higher throughput, I do think proceeding directly to a combinatorial library of the language model-proposed mutations could be quite effective.