I start from the Nipah virus attachment protein G (NiV-G) in complex with its human receptor (2VSM). I clean the NiV-G chain and identify a surface epitope in the receptor-binding site based on solvent exposure and proximity to ephrin-B2.

Using this epitope and the NiV-G structure as conditioning input, I generate diverse de novo binder sequences with BoltzGen. These sequences are then folded as target–binder complexes with Boltz2, so that the model can propose a full 3D interface for each design.

I evaluate the resulting complexes with simple geometric filters (inter-chain distance and contact count) and then score them with ipSAE (IPSAE repo) applied to the Boltz2 PAE and coordinates. Designs are clustered to avoid redundancy, and I submit the highest-ipSAE representative from each cluster, keeping only binders below 250 aa and removing near-duplicates within my own set.