TLDR:

1. To increase the success rate of de novo binders in expression and stability, we hypothesized that reusing validated proteins would help us reliably avoid these common failure modes.
2. We monitored specificity of our binders by discarding binders that confidently co-foldind to itself or to human sialidase (strcuturally similar, different seqeunce) (https://github.com/profdocpizza/Boltz2_IpSAE).
3. We interrogated some of our designs with molecular dynamics simulations in order to assess the stability of the complex dynamically. We ran the default protocol implemented in the drMD (https://github.com/wells-wood-research/drMD) molecular dynamics package, with three replicas per complex. The production simulations were run for 100 ns each. The metrics used to assess the stability of the designs were the RMSD and RMSF of the receptor, binder and complex, the solvent accessible surface area (SASA) of the complex, the ΔSASA (equal to the SASA of the complex minus the SASA of the receptor and the SASA of the binder), the number of contacts formed between receptor and binder (with a cutoff of 4.5 A), and the minimum distance between receptor and binder.
4. We ranked our designs by rosetta_total_binder, rosetta_total, rosetta_interaction_score, predicted_usability acquired via DE-STRESS (https://github.com/wells-wood-research/de-stress).

Detailed description:

Designs 1-6 We used miniproteins that are easily expressed and highly stable within a custom Boltz2–solubleMPNN iterative redesign cycle in the context of the Nipah G protein. We ran 10 generations, with 5 designs per generation. Each generation was derived from the best-performing backbone. Across generations, initial non-binders rapidly became confident binders, achieving ipSAE values up to 0.87 and ipSAE_min up to 0.84, while retaining their overall structural topology. We hope to see a higher success rate in expressibility and stability compared to fully denovo approaches. We adapted code from ProteinHunter to run Boltz and solubleMPNN.

Design 7 Additionally, we subjected a nanobody (PDB: 8XPS) to this iterative refinement pipeline. The nanobody is already known to bind the Nipah G protein. Interestingly, it often migrated away from its original binding site and instead gained affinity for the central region of the propeller (reflected by a substantially higher ipSAE).

Design 8-9 Finally, we also passed several BindCraft and ProteinHunter generated designs through this iterative refinement pipeline. We optionally ran BindCraft and ProteinHunter in a custom "sequence suggestion" mode where we add known seqeunces to the initial guess. We used natural binder loops (8xc4: QYYYSGSYPTPHN, 3d12: EYXPNLWGXE, 6vy5:QXLM ,8k3c:VXPXPRGY,2vsm:KXQEFSPNLWGXE) that interact with center of beta propeller as seqeunce suggestions and it often resulted with a similar binding motif.

Design 10 Rather than attempting to outperform nature, we aimed to work with it. We fused our binder sequences to a known strong binder (PDB: 8XPS) using multiple G₄S linkers. Because these binders target distinct sites on the Nipah G protein, the domains are expected to act synergistically.