not surprising this works, but it seems to improve Boltz2 IPTM for minibinders. another option is to simply use your favorite helix bundle or alpha solenoid as a scaffold or SS constraint

Pretty interesting that AFAICT the filtering was done after the fact (so, library 1 had no filtering). This could make it an excellent dataset for training/testing filters/rankers. Too bad it looks like the dataset is not public

It's a very cool result but IMO there are caveats. Inference is (mostly) slower. There is existing work on faster models (e.g. MiniFold or protenix mini), and also existing work on ensemble prediction. I doubt this works without training on AFDB, which bakes in inductive bias from triangle layers

could be fun to implement this in github.com/escalante-bi... -- and then try swapping AF2 for Boltz or Protenix, adding an inverse folding term during design, etc

mpnn actually improved the hit rate and affinities for af2 -- motivation same as af2cycler 😀. I wouldn't read much into boltz results, there are a ton of hyperparameter choices for boltz we made (e.g. # of diffusion steps, MSAs vs templates, refolding with AF2 multimer) that complicate the story

For these designs we used the proven BindCraft filters. `mosaic` is similar to BindCraft but supports many more models (AF2, Boltz 1+2, Protenix, several PLM and inverse folding models, any JAX model). We’re really excited about `mosaic` for more complex design objectives like dual target binders.

For this target it should be, I've run it on an A10 with 24gb

Try it out yourself with a demo marimo notebook: github.com/escalante-bi...

Still a work in progress; feedback and PRs welcome :)

p.s. this is a very small target but each one of these designs takes only 14 seconds to generate on an H100 (slightly longer to get discrete sequences)

Interesting! Seems very similar to what @sokrypton.org does with his contact losses; probably encourages all mass in one PAE bin? I added it to `boltz-binder-design` for boltz-1 and AF2: github.com/escalante-bi...

Sure, would love to! Just submitted a PR