Multi-Track De Novo Binder Design for RBX1

Team: Jozef Fülöp Affiliation: UCT Prague Contact: fulopj@vscht.cz

Background RBX1 (108 aa; UniProt P62877) contains an intrinsically disordered N-terminal segment (residues 1-39) and a zinc-coordinated RING-H2 domain (residues 40-108). To address this combination of disorder and rigid structure, we used three complementary computational design tracks and a shared scoring pipeline. Method We generated 1,748 unique binder candidates across three tracks. First, BoltzGen v0.3.1 (Stark et al., 2025) in sequence-only co-folding mode produced 1,329 designs by modeling binder and full-length RBX1 together. In this track, binders of 40-70 aa gave the best internal pass rate (22%). A separate hotspot-guided campaign centered on the 2LGV RING domain and the E2-binding groove did not yield passing designs (0/11,600). Second, BindCraft (Pacesa et al., 2025) generated 366 designs against the RBX1 RING domain (PDB 2LGV; residues 40-108) using AF2-based hallucination with E2-groove hotspot constraints. Third, ProteinMPNN/SolMPNN redesign of selected BoltzGen backbones yielded 53 additional sequences while keeping interface geometry fixed. Scoring and ranking All candidates were scored with ESM2 pseudo-log-likelihood and Boltz-2 co-fold ipTM. Boltz-2 coverage was 92%, and ESM2 coverage was 100%. A smaller subset was cross-checked with OpenFold3 (24 designs) and AF2-Multimer ipTM (366 BindCraft designs). Final ranking was driven primarily by Boltz-2 ipTM, with duplicate Boltz-2 predictions resolved by taking the minimum score as a conservative estimate. This ranking choice is consistent with recent binder-design benchmarking showing that interface-focused confidence metrics can improve prioritization performance, especially AF3-derived interface metrics (Overath et al., 2025). Related recent work from PXDesign also supports the value of confidence-based filtering and ranking across binder-design workflows (Ren et al., 2025). Selection We took the top 120 designs, applied diversity clustering (edit-distance threshold 0.2; maximum four designs per cluster), and retained 105. MMseqs2 novelty filtering against UniRef50, using both full-length sequences and 30-residue sliding windows at <=75% identity, reduced this to 102 passing candidates. The top 100 were submitted. In silico results All 100 submitted designs had Boltz-2 ipTM > 0.92 (mean 0.934; max 0.965). Lengths ranged from 40 to 120 aa. The final set included 58 BoltzGen designs, 40 BindCraft designs, and 2 MPNN redesigns. No top-ranked submitted design contained an unpaired cysteine. Across 300 predicted complexes, three recurring binding modes were observed: IDP+RING bridging (46%), IDP-dominant (44%), and RING-dominant (10%). In practice, the co-folding track was the main source of designs that spanned both the disordered N-terminus and the structured RING region. References

1. Stark H, Corso G, Barzilay R, Jaakkola T. BoltzGen: Toward Universal Binder Design. bioRxiv [Preprint]. 2025
2. Overath, M. D., Rygaard, A. S. H., Jacobsen, C. P., et al. (2025). Predicting Experimental Success in De Novo Binder Design: A Meta-Analysis of 3,766 Experimentally Characterised Binders. bioRxiv.
3. Pacesa, M., Nickel, L., Schellhaas, C., et al. (2025). One-shot design of functional protein binders with BindCraft. Nature, 646, 483-492.
4. Ren, M., Sun, J., Guan, J., et al. (2025). PXDesign: Fast, Modular, and Accurate De Novo Design of Protein Binders. bioRxiv.
5. Wohlwend, J., Corso, G., Passaro, S., et al. (2024). Boltz-1: Democratizing Biomolecular Interaction Modeling. bioRxiv.