Nipah Binder Competition

We computationally designed 2,000 candidate binders and ranked them using iPTM and PAE scores, selecting the top 10 designs after a ~23-hour GPU run. The final set consists of 8 antibody-based binders and 2 α-helical binders.

+5

We de novo design nanobodies using EAGLE (Epitope-specific Antibody Generation using Language-model Embeddings) - a method we developed in our lab. EAGLE is a diffusion-based deep-learning model that designs full-length nanobody and antibody sequences (both constant and variable regions) entirely in continuous language model (ESM) embedding space, without requiring predefined backbone structures.

+5

De novo nanobody-like and protein binders for the Nipah virus glycoprotein (NiV-G) were generated using BoltzGen, guided by a literature and MD-defined epitope.

+5

Modification of 1E5 with consistently positive prediction metrics in 50+ (each) AF3 recycles.

Why vote? These designs are engineered to be high-probability expressers and a test of whether a developability-first strategy can yield real binders.

+1

As a part of my master's project, I created a simple nanobody design pipeline consisting of: RFDiffusion for backbone design -> ProteinMPNN for sequence design -> Boltz-2 for structure design -> Rosetta Fast Relax -> ipSAE scoring and ranking. In this case, I created 1000 different backbones with 2 ProteinMPNN sequences each, leading to a total of almost 2000 designs.

+5

Approach We started from the pre-existing antibody–receptor complex structure (PDB: 6CMI), using its complete variable heavy and light chains as the design scaffold. Framework regions and overall CDR lengths were kept as close as possible to the original 6CMI antibody, while focusing redesign on the CDR loops.

To design VHH (nanobodies) against the Nipah virus glycoprotein G, a highly truncated version of the 2VSM structure was used as the antigen input for the public Germinal model. Germinal generates VHH designs using AF2-multimer hallucination jointly optimized with the antibody language model IgLM.

+5

We designed a library of 10,000 binders using state of the art diffusion, hallucination and pLM models (BindCraft, Latent X, La Proteina, ESM3, APM and DPLM2) and evaluated the designs using AlphaFold3 and Boltz2. Due to the discrepancy between the prediction methods (that is also reflected in the medium score of our collection even though we obtain high AlphaFold3 scores), we focused on AlphaFold3 for our ranking.

+5

We designed de novo minibinders against Nipah virus Glycoprotein G head domain using FoldCraft or BindCraft. To make design process faster we trimmed the head domain of Glycoprotein - 2VSM (337-511).

+5

TIMED Nanobody Workflow TIMED is a family of Convolutional Neural Networks (CNNs) models for protein sequence design. For this competition we used TIMED (vanilla), TIMED-Charge (charge aware), and co-TIMED (side-chain aware).

+3

This is a submission to the Nipah Binder Competition competition.

+5

This is a submission to the Nipah Binder Competition competition.

We strongly recommend the immediate manual selection and rigorous experimental evaluation of our highly refined portfolio of ten novel nanobody sequences, which were meticulously engineered based on robust, clinically validated structural scaffolds derived from the latest commercial and clinical-stage nanobody therapeutics, thereby ensuring inherent stability and optimal pharmacological profiles for genuine therapeutic applications; this focused engineering effort centered specifically on optimizing the three Complementarity-Determining Regions (CDRs)—the critical binding loops—to confer specific, high-affinity antigen recognition capabilities that advance beyond simple in silico homology screening. Crucially, the design methodology integrated the advanced generative power of Boltz-Gen for expansive sequence space exploration with the proprietary, in-house structure prediction model, DynaHelix, which is specifically optimized for enhancing prediction precision by pioneering the automated search across diverse potential interface configurations during the modeling process, thereby mitigating the limitations of static structural predictions and identifying the most viable binding geometry; despite the fact that the resulting CDR-antigen interaction interfaces might initially exhibit lower scores in standard metrics such as interface predicted Template Modeling score (iptm) or Predicted Aligned Error (PAE)—a common and expected outcome when generating novel sequences outside of known structural datasets—we maintain that the depth of the structure-centric validation and multi-model optimization warrants their selection, as it prioritizes functional feasibility and the probability of a therapeutic mechanism over raw predictive statistics; consequently, these ten structure-guided candidates offer truly novel binding modalities and structural solutions that demand expert appraisal and represent a significant strategic opportunity to accelerate the development of next-generation nanobody therapeutics.

+5

This is a submission to the Nipah Binder Competition competition.

+5

De Novo Nipah RBP Binders: Targeting Critical Hotspots via Boltzgen & IPSAE Validation

+5

I, Stephanie Reyes , along with the support of Javier Uzcátegui , performed the respective literature review and subsequently selected the key amino acids of Ephrin B3 through molecular docking. We worked according to the binding domain of Ephrin B3, specifically focusing on the surface amino acids of this ligand.

The idea is to leverage the NCBI Virus database to map variants of the Glycoprotein G to host organisms. It enables to build a virus-host paired MSA (sequences aligned with the Glycoprotein G and sequences aligned with Ephrin-B3 receptor with MAFFT).

+5

This is a submission to the Nipah Binder Competition competition.

Used Germinal to design scFv

This is a submission to the Nipah Binder Competition competition.