=== ENTRY DETAILS ===

- QR code with entry information is available, in the "Photos" section of this event page.
- Gate closes at 18:15 - no late entries.

In this session, we’ll walk through one of the newest structural biology foundation models: Boltz-2.

The model’s key achievement is its state-of-the-art ability to predict binding affinity - measuring how tightly small molecules bind to proteins.

This process is critical in drug discovery, and until now, machine learning models have struggled to deliver accurate and scalable predictions.

Boltz-2 achieves predictive performance on par with the best atomic-level simulation methods (e.g free-energy perturbation), while being an E2E deep learning model, and running ~1000x faster. Also, it is fully open source.

The session will be divided into 2 parts:
- 1) Intro to the domain of binding affinity prediction to understand the modeling problem and highlight its importance in drug discovery
- 2) A detailed overview of the model’s architecture, with a focus on Boltz-2’s unique design choices (affinity module, controllability) and how it differs from AlphaFold 3 and its predecessor, Boltz-1.