These binders target the cryptic pocket discovered in PDB: 8XPY. Boltz2 can't see this binding mode hence the low ipsae (and incorrect complex predictions!!). If you template AF3 with the target from 8XPY you can see the correct binding modes. I didn't mean to target a different spot than everyone else!! That's just the pdb they listed in the initial twitter post. I will admit that this cryptic pocket is way easier.

These designs were NOT tested experimentally in any capacity.

My design method looks a lot like Cao et. al. 2022 just with RFDiffusion instead of RifDock and Rosetta. I also hand-filtered the designs at several steps along the way (does this backbone look like it could work?)

Correct docks: https://drive.google.com/file/d/1cJFjN1RbdAcV84lpOptIQlj93JAgdUdT/view?usp=sharing