This method combines AI-assisted sequence ideation with mechanistic design principles and independent physics-based validation. The approach begins with de novo generation of long amphipathic α-helices (48–55 aa) containing an oriented aromatic motif (W/F/Y) positioned to target the conserved Trp125 hydrophobic pocket on NiV-G. The design intent is to produce rigid, membrane-surface “rider” helices that sterically occlude the Ephrin-B2/B3 binding site without inserting into or disrupting the host membrane.

Sequences are refined using LLM-assisted pattern optimization (charge distribution, helical periodicity, and aromatic spacing), followed by structural prediction with PEP-FOLD to confirm helix continuity. Candidate models are evaluated against the NiV-G pocket using PyMOL mapping, and binding modes are explored via HADDOCK2.4 with restraints applied only to NiV-G pocket residues to avoid sequence overfitting. Final selections are based on PRODIGY interface energetics, ΔG/Kd predictions, and reproducibility of apolar–apolar contact patterns relative to the native Ephrin–NiV interface.