Boltz-2 - Towards Accurate and Efficient Binding Affinity Prediction

Explore a groundbreaking structural biology foundation model in this 59-minute conference talk that demonstrates significant advances in predicting both biomolecular complex structures and binding affinity. Learn about Boltz-2's innovative controllability features, including experimental method conditioning, distance constraints, and multi-chain template integration for enhanced structure prediction capabilities. Discover how this model represents the first AI system to approach the performance of free-energy perturbation (FEP) methods in estimating small molecule-protein binding affinity while maintaining over 1000× greater computational efficiency. Examine the model's strong correlation with experimental readouts across multiple benchmarks and understand its practical applications in drug discovery workflows. See how coupling Boltz-2 with generative models for small molecules enables the identification of diverse, synthesizable, high-affinity binders, demonstrated through absolute FEP simulations on the TYK2 target. Gain insights into the open-source release of Boltz-2 weights, inference, and training code under a permissive license, designed to accelerate research at the intersection of machine learning and biology for both academic and industrial applications.