We designed de novo protein binders against human RBX1 using a multi-branch computational pipeline combining generative backbone design, parallel sequence design, and structure-based complex screening.

To guide interface placement on the RBX1 RING domain, we leveraged prior knowledge that glomulin is a natural RBX1 inhibitor that blocks E2 recruitment by binding the RING surface. In particular, we used reported RBX1 interface “hotspot” residues W33, W35, I44, R46, W87, and R91 as reference positions when defining the targeted binding region and when prioritizing designs that make favorable contacts near these sites.

Candidate binder backbones were generated with RFdiffusion3, and sequences were co-designed using ProteinMPNN as part of the same design stage (RFdiffusion3 + ProteinMPNN). In parallel, we carried out independent sequence design/optimization using Protenix-Design and ODesign, producing complementary candidate sets and improving sequence/structural diversity.

All candidates were then evaluated with AlphaFold3 by predicting RBX1–binder complex structures. Designs were filtered and ranked based on predicted complex confidence and interface quality , with additional emphasis on predicted interactions concentrated around the glomulin-inspired hotspot region. The top-ranked sequences were selected for submission while satisfying the competition constraints on de novo design, length, and sequence novelty.