Boltz-2: An Open Source Revolution in Predicting Molecular Interactions

On June 6, 2025, the Massachusetts Institute of Technology (MIT) and the biotechnology company Recursion announced the release of Boltz-2, an open-source artificial intelligence (AI) model capable of simultaneously predicting the three-dimensional structure of biomolecules and their binding affinity. This breakthrough promises to transform drug discovery by significantly accelerating the process of selecting potential therapeutic compounds.

A Major Technological Breakthrough for Drug Discovery

Boltz-2 is the first biomolecular co-folding model to combine structure prediction and binding affinity, achieving accuracy close to physics-based free energy perturbation (FEP) calculations, but with speeds up to 1000 times faster in standard benchmarks. Developed jointly by researchers from MIT and Recursion, and trained on Recursion's BioHive-2 supercomputer, Boltz-2 represents a significant advancement over previous models such as AlphaFold3 and Boltz-1.

According to Regina Barzilay, Distinguished Professor of AI and Health at MIT, "Boltz-2 not only solves this crucial problem, but also helps scientists discover new biological insights and ask questions they couldn't before with more computationally intensive standard approaches."

The model was trained on approximately 5 million binding affinity measurements, molecular dynamics simulations, and extensive distillation data, enabling it to accurately predict interactions between molecules and their biological targets.

Democratizing AI for Biomedical Research

One of the most remarkable aspects of Boltz-2 is its availability as open source under the MIT license, including the model code, weights, and training pipeline, for both academic and commercial use. This openness aims to make AI tools accessible to drug developers, especially those without the resources to develop their own models.

Najat Khan, Chief R&D Officer and Chief Commercial Officer at Recursion, emphasizes that "selecting the right molecules early is one of the most fundamental challenges in drug discovery, with implications for the success or failure of R&D programs." By enabling rapid and accurate prediction of structure and binding affinity, Boltz-2 provides R&D teams with a powerful tool to effectively prioritize the most promising compounds.

The model also offers advanced features such as "Boltz-steering," which enhances the physical plausibility of predictions, and user control options through templates, methods, and contact conditioning. These features allow researchers to tailor the model to specific types of molecules, making Boltz-2 even more powerful as a tool to accelerate discovery.

Boltz-2 represents a major advancement in the application of AI to drug discovery, combining high accuracy with unprecedented speed and scale. By making this model accessible to the scientific community, MIT and Recursion are fostering increased collaboration and accelerated innovation in the field of biomedical research. This initiative could well mark the beginning of a new era where AI plays a central role in developing more effective and faster treatments for patients.