Sequence design: – De novo generation of amphipathic helices (48–55 aa) using LLM-assisted optimization. – Constraints: high helical propensity (A/L/E/K/Q-rich), amphipathic surface patterning, centrally positioned aromatic cluster (W–L–Y–F). – Filtered to avoid natural-sequence similarity.

Structure prediction: – PEP-FOLD 3.5: top 10 models inspected for helix continuity, absence of kinks, and terminal disorder.

Pocket mapping: – PyMOL visualization of 2VSM; target residues: Cys238–241, Arg242, Ile401, Arg402, Gln490–Ser491, Trp504, Val507.

Docking: – HADDOCK 2.4 with active residues defined only on NiV-G (no restraints on peptides). – Rigid/semi-flexible refinement clusters analyzed.

Interface scoring: – PRODIGY ΔG and Kd prediction. – Contact decomposition (apolar/apolar, apolar/polar, charged interactions) compared to native Ephrin-B2/NiV-G interface. – Selection criteria: nanomolar-range predicted affinity, deep or shallow occlusion of Trp125 pocket, stable helical architecture.