Key Takeaways
• Boltz-2 delivers AI-driven binding affinity predictions 1000x faster than traditional free energy methods—dramatically accelerating virtual screening workflows.
• Jointly developed by MIT and Recursion, Boltz-2 predicts both 3D structure and binding affinity—a leap beyond AlphaFold3 in molecular modeling precision.
• Fully open-source under the MIT License, Boltz-2 democratizes next-gen AI for biopharma, unlocking customizable, scalable R&D pipelines.Key Takeaways
• Boltz-2 delivers AI-driven binding affinity predictions 1000x faster than traditional free energy methods—dramatically accelerating virtual screening workflows.
• Jointly developed by MIT and Recursion, Boltz-2 predicts both 3D structure and binding affinity—a leap beyond AlphaFold3 in molecular modeling precision.
• Fully open-source under the MIT License, Boltz-2 democratizes next-gen AI for biopharma, unlocking customizable, scalable R&D pipelines.
AI-First Drug Discovery: Joint Modeling for Next-Level Screening
Boltz-2 is a next-gen biomolecular foundation model co-developed by MIT and Recursion, designed to predict both molecular structure and binding affinity with unmatched efficiency. Unlike traditional models that separate structure prediction (e.g., AlphaFold3) and affinity estimation, Boltz-2 unifies both in a single, fast AI pass. This joint modeling capability enables real-time, high-accuracy virtual screening—critical for accelerating hit-to-lead selection and reducing R&D costs in early-phase discovery.
1000x Faster than FEP: Speed That Changes Strategy
Free energy perturbation (FEP), once the gold standard for affinity prediction, is now outpaced by Boltz-2’s AI architecture. The model achieves near-parity with FEP accuracy but operates up to 1000x faster, making it feasible to screen vast compound libraries in hours, not weeks. With drug attrition often linked to early selection errors, this breakthrough enables R&D teams to prioritize better leads, sooner—transforming virtual screening from a bottleneck to a competitive advantage.
Physical Plausibility Meets AI Control: Smarter Predictions, More Insight
Boltz-2 incorporates physics-aware AI tools like Boltz-steering and template-based conditioning, delivering more interpretable and customizable predictions. It also models structural dynamics, such as B-factors, to capture nuanced molecular behavior. These features help researchers simulate more realistic interactions, ask deeper biological questions, and reduce the trial-and-error common in AI-driven design—accelerating precision medicine pipelines.
Open Innovation at Scale: CASP16 Leader, Global Access
Trained on 5M+ binding affinity assays and advanced molecular dynamics data, Boltz-2 led the CASP16 affinity prediction challenge. Its full release includes source code, weights, and training workflows—allowing academic labs, startups, and pharma enterprises to embed it into their platforms immediately. The model is open-source under the MIT License, representing a shift toward democratized AI tools that scale from bench to enterprise without IP restrictions.
About MIT CSAIL and Jameel Clinic
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is one of the world’s leading research institutions in computing, AI, and machine learning. Its work spans foundational AI research to applications in healthcare, robotics, and life sciences. The Jameel Clinic for Machine Learning in Health at MIT focuses on developing AI tools that transform disease detection, drug discovery, and medical understanding.
About Recursion
Recursion is a clinical-stage biotechnology company pioneering the use of artificial intelligence, automation, and data science to industrialize drug discovery. With one of the largest proprietary biological and chemical datasets in the industry, Recursion integrates deep learning across its platform to accelerate therapeutic development for complex diseases.
By