We present a two-track de novo peptide binder design pipeline targeting the E2-recruitment surface of RBX1, focusing on the N-terminal region (residues 18–36) that mediates interaction with GLMN in the CUL1-RBX1-GLMN complex (PDB: 4F52). This interface, characterized by a coupled folding-and-binding beta-strand insertion, presents 32 hydrogen bonds and 5 salt bridges across 17 residues, with core hotspots at ARG21, PHE22, TRP27, LEU32, and TRP33 (total buried surface area ~1,904 Ų).

Track 1: pepMLM (71 sequences, 30 aa): We used pepMLM, a masked language model fine-tuned on therapeutic peptides, to generate candidates conditioned on the RBX1. From an initial pool of 500 generated sequences, we applied a multi-stage filter: pseudo-perplexity < 30, standard amino acids only, no poly-repeat runs (≥4 identical consecutive residues), and sequence diversity ≥ 35%. This yielded 212 candidates, which were scored using PeptiVerse — a unified platform for therapeutic peptide property prediction — evaluating binding affinity against RBX1, solubility and hemolysis, and serum half-life. The top 71 candidates by composite score (weighted: binding 45%, solubility 25%, hemolysis safety 20%, half-life 10%) were selected, all achieving medium or tight predicted binding (pKd ≥ 7.0).

Track 2: moPPIt (29 sequences, 25 aa): We used moPPIt, a multi-objective discrete flow matching model for motif-specific peptide binder generation, targeting the RBX1 N-terminal hotspot motif directly. All 29 generated candidates were scored on PeptiVerse and included in the final submission. This track produced 14 tight-binding candidates (pKd ≥ 9.0), including the top-ranked binder (pKd = 9.761).

Candidate selection and ranking: All 100 sequences were ranked by predicted binding affinity (pKd) against the full RBX1 sequence. The final submission contains 21 tight-binding (pKd ≥ 9.0), 41 strong medium-binding (pKd 8–9), and 38 medium-binding (pKd 7–8) candidates. No weak binders (pKd < 7) were included. All sequences are de novo generated with no motif scaffolding, consist of standard amino acids, and are ≤ 30 residues in length.