Computational Design of High-Affinity Binders Targeting the RBX1 RING Domain

Abstract

We present a de novo computational pipeline for designing protein binders against the RING domain of RBX1. Our approach integrates structural biology insights with a multi-agent cognitive synergy framework (ARCHAOS-ReSEADCaVe) to generate and screen candidate sequences. From an internal library of over 1500 candidates, we have curated a focused set of 3 high-quality sequences for submission, representing the pinnacle of our design strategy.

1. Background & Target Analysis

Target: RBX1 (RNF75, UniProt: P62877). Its RING domain (approx. residues 40-108) is essential for recruiting E2 ubiquitin-conjugating enzymes. Key Structural Challenges:

1. Zinc-Coordination: The RING domain is stabilized by two zinc ions (PDB: 3DPL).
2. Native Protein-Protein Interfaces: The domain has conserved interfaces for binding Cullin proteins (e.g., Cul1, PDB: 1LDK) and E2 enzymes (PDB: 3DPL).
3. Disordered N-terminus: The N-terminal region is predicted to be intrinsically disordered (AlphaFold2 DB: AF-P62877-F1). Design Goal: To design novel, high-affinity binders targeting accessible regions of the RBX1 RING domain while strictly avoiding zinc-coordinating residues and native interaction sites.

2. Method: A Cognitive Synergy Pipeline

Our design strategy was executed by a coordinated ensemble of specialized AI agents.

2.1 Target Cognizance & Constraint Definition

• Epitope Selection: The α2 helix (residues 86-94) and its flanking regions were identified as primary targets.
• Quantitative Constraint Encoding: Rules were translated into generative constraints, such as obligatory polar networks and strict zinc exclusion.

2.2 Synergistic Generation & Evaluation Loop

1. Backbone Generation: Performed using RFdiffusion.
2. Sequence Design: ProteinMPNN (with SolubleMPNN bias) was used.
3. Rapid In-silico Funnel: All designs were filtered by AlphaFold2 pLDDT (≥82) and HDock.
4. Deep Evaluation: Top candidates were evaluated using Rosetta for ΔG calculation.
5. Strategic Analysis & Iteration: Results were analyzed and fed back as refined constraints.

2.3 Final Filtering & Library Curation

From over 1500 generated candidates, we applied a rigorous multi-objective Pareto optimization to select the final 3 sequences for submission. This curation ensures that only the highest quality, most promising binders are presented for experimental validation.

3. In-silico Results & Summary

• Submitted Library Size: 3 de novo protein sequences.
• Sequence Length: 110-130 amino acids per sequence.
• Predicted Quality: All submitted sequences exhibit high pLDDT (>89), excellent predicted expressibility, and strong binding affinity metrics.
• Innovation: Our pipeline's core innovation lies in the cognizant, closed-loop synergy.

4. Submitted Sequences Highlight

• Rank 1 (GNYDY...): Features a unique π-helix element and a tripartite polar cluster, designed for a novel binding mode targeting the α2-α3 loop.
• Rank 3 (DHYM...): Showcases a balanced hydrophobic/aromatic interface with a predicted high-affinity interaction profile.
• Rank 6 (SQKP...): Offers an alternative scaffold with stabilizing disulfide motifs, demonstrating design diversity.

5. References

1. UniProt Consortium, UniProt: P62877 (RBX1).
2. AlphaFold Protein Structure Database, AF-P62877-F1.
3. Zheng et al. (2002). Nature. Structure of the Cul1–Rbx1–Skp1–F boxSkp2 SCF complex (PDB: 1LDK).
4. (PDB References: 3DPL, 1U6G, etc.)