Title: HotBind-RBX1: Hotspot-Guided De Novo Peptide Binder Design against RBX-1 Using BindCraft and Multi-Stage Docking Validation Team: Rik Ganguly (Junior Engineer, Mphasis AI & Applied Tech Lab at Ashoka, Ashoka University), Siddhant Poudyal (Project Research Scientist-I, Ashoka University), Gautam Ahuja (Multimodal Biomedical Data Scientist & ML Engineer, Department of Computer Science, Ashoka University)

Background: RBX-1 (RING Box Protein 1, UniProt P62877) is a 108-residue E3 ubiquitin ligase component of the SCF complex, critically involved in cell cycle regulation and oncogenesis. Its C-terminal RING-H2 domain, stabilised by three zinc ions, presents a structurally rigid yet functionally dynamic surface that is challenging for standard binder design methods. We targeted this domain using a structure-guided hotspot identification strategy followed by deep-learning-based de novo peptide design.

Approach: The HotBind-RBX1 pipeline proceeds in four stages. (1) Binding pocket identification: CASTpFold was used to identify a well-defined surface cavity (Pocket ID 1; 26 residues) on the RING-H2 domain. Evolutionary conservation analysis via MSA of RBX-1 orthologs identified eight highly conserved functional residues - TRP35, ASP36, ASP40, ARG46, ASN47, TRP72, HIS80, TRP87 - confirmed as a high-frustration cluster by Frustratometer energetic profiling. These eight residues were designated as design hotspots. (2) De novo binder design: BindCraft was used in two parametric runs targeting the hotspot residues. Run 0 (flexible protocol, 20–40 aa, Peptide filter) produced 25 candidate sequences; Run 1 (fixed protocol, 6–24 aa, Peptide_Relaxed filter) produced 3 pre-folded complexes. All sequences were folded via ESMFold where structures were not directly provided. (3) Docking validation: All 28 (24+3) candidates were docked against the RBX-1 structure using both HADDOCK 2.4 (web server) and HADDOCK 3 (local CLI), enabling cross-pipeline validation of binding poses and scores. (4) Affinity estimation and ranking: PRODIGY was applied to all HADDOCK-refined complexes to predict binding free energy (dG, kcal/mol) and dissociation constant (Kd). Final submission ranking used a weighted composite score integrating PRODIGY affinities (40%), HADDOCK scores (27%), cluster size and SD (13%), BindCraft dG and binder energy score (13%), and pLDDT/pTM (7%). Sequences flagged as low-complexity by Shannon entropy analysis were deprioritised.

In-silico results: 28 sequences passed all BindCraft filters and both docking pipelines. Predicted Kd values (PRODIGY on HADDOCK 2.4 complexes) range from 71 fM to 760 nM, with the top-ranked binder (RBX1_l38_s814322_mpnn1, 38 aa) achieving dG = -12.1 kcal/mol and Kd = 1.2 nM. Novelty was confirmed by MMseqs2 and DIAMOND searches against UniRef50 - zero sequences exceeded 75% identity to any known protein, with all DIAMOND hits non-significant (E-value > 600).