RBX1 (RING box protein 1) is a 108-amino-acid component of the cullin-RING E3 ubiquitin ligase (CRL) complex, which represents the largest family of E3 ligases in humans. CRLs regulate the degradation of approximately 20% of cellular proteins and thereby control cell-cycle progression, DNA repair, and signal transduction. RBX1 functions as the catalytic subunit by recruiting E2 ubiquitin-conjugating enzymes. Accordingly, the design of protein binders that target RBX1 and disrupt its function may inhibit CRL-mediated protein degradation and thus holds potential for cancer therapy. RBX1 exhibits a pronounced bipartite structural architecture. Residues 1-39 form the N-terminal region, which lacks stable secondary structure and is therefore highly disordered and flexible. In contrast, residues 40-108 (DIVVD...FQKYGH) constitute the core catalytic engine of RBX1, namely the RING-H2 finger domain. This region adopts a highly compact fold and coordinates three structural zinc ions, conferring substantial conformational rigidity. Within this compact domain, the zinc-coordination network stabilizes multiple beta-strands and the central alpha-helices (alpha 1 and alpha 2). The zinc-coordinating residues are Cys42, Cys45, Cys53, Cys56, Cys68, Cys75, Cys83, Cys94, His77, His80, His82, and Asp97 [1]. During the backbone-generation stage, zinc-coordinating residues were excluded from hotspot definition. At the C-terminus of RBX1, E2 recognition is mediated primarily by a hydrophobic residue network located within the central loop region of the RING domain. A review of the literature indicated that the principal residues contributing to RBX1-E2 binding are Trp87, Arg91, Ile44, Leu96, Pro95, Ala43, and Arg46, among which Trp87 and Arg91 appear to be the most critical [1-3]. On this basis, four hotspot sets were defined: set 1, Ile44/Trp87/Arg91/Leu96; set 2, Ile44/Trp87/Arg91/Pro95/Leu96; set 3, Ile44/Arg46/Trp87/Arg91/Leu96; and set 4, Ile44/Arg46/Trp87/Arg91/Pro95/Leu96. Using the RBX1 structure with Protein Data Bank accession number 2LGV as the target, RFdiffusion3 was applied to generate a total of 500 binder backbones across the four hotspot sets. The generated backbones were then subjected to sequence redesign using LigandMPNN, with three sequences generated per backbone. Preliminary filtering based on overall_confidence and ligand_confidence retained 718 sequences (sequence version 1). The corresponding structural files (packed_pdb) for these 718 sequences were then input into Rosetta and relaxed using the FastRelax protocol, with full relaxation of the binder backbone and side chains and limited adjustment of side chains in the target protein interface region, yielding the first relaxed backbone set (backbone version 1). At this stage, the local geometry of the binder backbone became more physically realistic. These backbone version 1 structures were then returned to LigandMPNN for a second round of sequence redesign, with two sequences generated per backbone. After re-filtering based on overall_confidence and ligand_confidence, 469 sequences (sequence version 2) were retained. The corresponding structures were subsequently subjected to the same Rosetta FastRelax protocol to generate backbone version 2. These backbone version 2 structures then underwent an additional round of sequence redesign with LigandMPNN, and 300 sequences were ultimately advanced to the structure-prediction stage. Structure prediction was performed using AlphaFold3 (AF3), after which pae_interaction and plddt_binder were calculated for each model. Finally, the AF3-predicted structures were input into Rosetta for relaxation and dG_separated calculation. Using pae_interaction < 10, plddt_binder > 89, and dG_separated <= -30 as the selection criteria, 33 sequences were ultimately retained.