Introducing Boltz-1: Democratizing Biomolecular Interaction Modeling

Today, the MIT Jameel Clinic is thrilled to announce the release of Boltz-1, an open-source model designed to accurately model complex biomolecular interactions. Boltz-1 stands as the first fully commercially available open-source model to achieve AlphaFold3-level accuracy in predicting the 3D structure of biomolecular complexes. This marks a significant step forward in democratizing access to advanced biomolecular modeling. By releasing training and inference code, model weights, and training data under the MIT license, we aim to establish Boltz-1 as a modeling backbone for researchers worldwide, setting a new standard in open-source structural biology.

Performance

We evaluate the performance of Boltz-1 against Chai-1, the first closed-sourced but publicly available replication of AlphaFold3,  on various datasets and we conclude that Boltz-1 matches the performance of Chai-1 and, therefore, AlphaFold3. For example, when evaluated on CASP15, Boltz-1 demonstrates particularly strong protein-ligand and protein-protein performance achieving an LDDT-PLI of 65%, compared to 40% for Chai-1, and a proportion of DockQ>0.23 of 83% vs 76% of Chai-1.

For a full overview of results, architectural advancements, and implementation details, please see our technical report.

Fostering Open Science

This open-source release aims to empower researchers and organizations around the world to experiment and innovate with Boltz-1. We envision Boltz-1 as a foundation upon which researchers can build, collaboratively advancing our collective understanding of biomolecular interactions and accelerating discoveries in drug design, structural biology, and beyond.

We invite you to try out the model our GitHub repository and connect with us on our Slack channel — to discuss advancements, share insights, and foster collaboration around Boltz-1.

A lot of work went into Boltz-1 and we are proud to be the first academic institution to develop a model with this level of accuracy. We are particularly grateful to the support of the MIT Jameel Clinic; Genesis Therapeutics for ML engineering, infrastructure, and computational support; as well as the U.S. Department of Energy for their computational support.

This work was also supported by the NSF Expeditions grant (award 1918839: Collaborative Research: Understanding the World Through Code); the DTRA Discovery of Medical Countermeasures Against New and Emerging (DOMANE) Threats program; and the MATCHMAKERS project supported by the Cancer Grand Challenges partnership financed by CRUK (CGCATF-2023/100001) and the National Cancer Institute.

The Future of Boltz-1

Boltz-1’s open-source release is an exciting step forward, but we’re just getting started. We have major improvements in the works to enhance its capabilities in modeling complex interactions, and we plan to release these in the coming months! Stay tuned!

— Jeremy Wohlwend, Gabriele Corso and Saro Passaro

On behalf of the whole Boltz-1 team: Jeremy Wohlwend, Gabriele Corso, Saro Passaro, Mateo Reveiz, Ken Leidal, Wojtek Swiderski, Tally Portnoi, Itamar Chinn, Jacob Silterra, Tommi Jaakkola and Regina Barzilay