BoltzGen: Toward Universal Binder Design

Area of Science:

• Computational biology
• Protein engineering
• Artificial intelligence in drug discovery

Background:

• Designing novel proteins and peptides with specific binding capabilities is crucial for therapeutic and diagnostic applications.
• Existing methods often struggle with complex target structures and diverse binder modalities.
• Integrating structural reasoning into generative models is key for accurate target-binder interaction prediction.

Purpose of the Study:

• To introduce BoltzGen, an all-atom generative model for designing proteins and peptides across all modalities to bind a wide range of biomolecular targets.
• To enable precise control over the design process using a flexible specification language.
• To experimentally validate the model's performance in diverse wet-lab campaigns.

Main Methods:

• Developed an all-atom generative model (BoltzGen) unifying protein design and structure prediction.
• Implemented a flexible design specification language for controlling covalent bonds, structure constraints, and binding sites.
• Conducted eight diverse wet-lab design campaigns with 26 targets, including nanobodies, peptides, disordered proteins, and small molecules.
• Validated 15 nanobody and protein binder designs against nine novel targets.

Main Results:

• BoltzGen demonstrates strong structural reasoning for target-binder interactions.
• Achieved state-of-the-art protein folding performance within the generative model.
• Generated nanomolar binders for 66% of targets across nanobody and peptide modalities against novel targets.
• Experimental validation confirmed functional binders for diverse targets and modalities.

Conclusions:

• BoltzGen represents a significant advancement in AI-driven protein and peptide design.
• The model's ability to unify design, prediction, and experimental validation accelerates the discovery of novel binders.
• The open-source release of code and data facilitates further research and application in biomolecular design.