To design binders for RBX1, I decided to use BoltzGen to design initial binder scaffolds, which would be later optimized by an ensemble of solMPNN and ThermoMPNN. For the BoltzGen binder scaffolds, I used the "protein-anything" design protocol with inverse folding, budgeting for the top 10 of 1,000 binder designs. These designs were then validated and filtered for ipTM > 0.7 by AlphaFold. Following filtering, I optimized the non-interfacing binder residues with the ProteinMPNN's soluble model, and orthogonally validated the resulting sequences with AlphaFold, taking the top-scoring MPNN sequence if it maintained an ipTM > 0.7. Finally, to optimize binder affinity and interfacing residues, I used ThermoMPNN and kept the top 5 most stabilizing (negative ddG) point-mutants.