This method is organized around the idea that RBX1 should be evaluated across several structurally distinct binding contexts rather than through a single reference pose. The sequence set therefore covers the 4F52 main-face state, functional-state conformations such as 8Q7R and 8PQL, and the orthogonal 2LGV display-state geometry, while also tracking additional contact-supported placements in structures such as 3RTR and 4P5O when refolding recovers the same interface logic outside the primary state.

  The design strategy combines four complementary workflows:   • BindCraft de novo mini-proteins: Establish a seed-free main-face binding baseline.   • BindCraft + ProteinMPNN redesign: Utilize BindCraft backbones redesigned with ProteinMPNN to preserve robust geometries while enhancing sequence quality and foldability, contributing to both main-face and functional-state binders.   • Proteina + ProteinMPNN redesign: Employ Proteina-seeded scaffolds refined by ProteinMPNN to expand the structural diversity beyond standard BindCraft solutions. These play a primary role in targeting the 2LGV display-state, alongside selected 4F52 entries.   • EvoBind: Generate peptide-format probes for the 2LGV and 8Q7R states to explore shorter, orthogonal binding geometries that folded mini-proteins might not efficiently access.

  This dataset was assembled based on the following three related hypotheses: First, RBX1 may admit compact folded binders on the 4F52 main-face surface. Second, functional-state conformations may expose additional tractable interfaces that should be preserved near the front of the sequence set. Third, display-state and peptide-format routes may reveal RBX1-binding solutions that are structurally different from the leading folded mini-protein entries and therefore worth carrying forward even when their interface scale or topology differs from the main-face binders.

  Selection emphasizes structural evidence and developability, with additional weighting from interface scale, cross-state support, novelty preference, and route-aware calibration. Candidates were prioritized when they formed interpretable contact patches, reached a meaningful set of RBX1 residues, recovered credible monomer folds, and, where possible, retained compatible contact-supported placements in more than one RBX1 conformation. Broad-looking surfaces were discounted when the underlying contact footprint was too small, and peptide entries were calibrated separately so that narrow single-state signals would not dominate the final ranking.

  Overall, the final 30-sequence set favors state coverage, scaffold breadth, and interpretable binding logic. The full project rationale, figures, and supporting materials are documented at:https://github.com/tuna010101/RBX1-Binder-Design-Portfolio