Researchers from the Massachusetts Institute of Technology (MIT) and biotech company Recursion (NASDAQ: RXRX) have announced the open-source release of Boltz-2, a breakthrough AI model that dramatically accelerates biomolecular structure prediction and binding affinity calculations for drug discovery. The next-generation model represents a significant advancement in computational drug discovery, offering unprecedented speed while maintaining high accuracy in virtual screening applications.
Revolutionary Speed and Accuracy Combination
Boltz-2 distinguishes itself as the first open-source biomolecular co-folding model capable of jointly predicting 3D complex structures and molecular binding affinity. In standard benchmarks, the model approaches the accuracy of free energy perturbation (FEP), a gold-standard physics-based method used throughout the pharmaceutical industry, while performing up to 1,000 times faster.
"Accurately predicting how strongly molecules bind has been a long-standing challenge in drug discovery—one that required novel machine learning and computer science techniques to address," said Regina Barzilay, MIT School of Engineering Distinguished Professor for AI and Health, AI faculty lead at Jameel Clinic and CSAIL principal investigator. "Boltz-2 not only addresses this crucial problem but also helps scientists uncover new biological insights and ask questions they couldn't before with standard approaches that are more computationally intensive."
The dramatic speed improvement significantly lowers the cost and time required for large-scale molecular screening, directly addressing a critical bottleneck in small molecule discovery that has long plagued pharmaceutical research and development.
Technical Innovations and Performance Benchmarks
Built on the foundation of its predecessor Boltz-1 and inspired by models like AlphaFold3, Boltz-2 incorporates several key technical differentiators that set it apart from existing biomolecular prediction methods. The model was trained using Recursion's NVIDIA-powered supercomputer, BioHive-2, utilizing expanded datasets that include molecular dynamics simulations and approximately 5 million binding affinity measurements.
The model demonstrates superior predictive power across multiple benchmarks, outperforming all CASP16 affinity challenge participants. Boltz-2 achieves near-FEP accuracy on the widely adopted FEP+ benchmark while being over 1,000 times faster and less computationally expensive than traditional physics-based approaches.
Among its advanced capabilities, Boltz-2 uniquely models 3D protein-ligand complexes while simultaneously estimating dynamics like B-factors. The model also incorporates "Boltz-steering" to enhance physical realism and offers customizable outputs through template, contact, and method conditioning for targeted predictions.
Industry Impact and Accessibility
The development represents a collaborative effort combining MIT's academic expertise with Recursion's AI research capabilities and computational infrastructure. Najat Khan, Chief R&D Officer and Chief Commercial Officer at Recursion, emphasized the model's potential impact on pharmaceutical development.
"Selecting the right molecules early is one of the most fundamental challenges in drug discovery, with implications for whether R&D programs succeed or fail," Khan said. "By predicting both molecular structure and binding affinity simultaneously with unprecedented speed and scale, Boltz-2 gives R&D teams a powerful tool to triage more effectively and focus resources on the most promising compounds."
In line with MIT and Recursion's commitment to advancing the field through transparency and accessibility, Boltz-2 is released under the permissive MIT license. The open-source release includes the full model weights and training pipeline, enabling scientists to easily adapt the model for specific types of molecules and making it accessible for both academic and commercial use.
Advancing Virtual Screening Capabilities
The model's ability to perform large-scale and accurate virtual screening makes previously impractical computational approaches more feasible for drug discovery teams. This advancement addresses the fundamental challenge of molecular selection in early-stage drug development, where the identification of promising compounds can determine the success or failure of entire research programs.
Barzilay noted that the open-source nature of Boltz-2, including its training code, allows scientists to customize the model for specific molecular types, potentially making it even more powerful as a discovery acceleration tool. The model's development was led by the Boltz team at MIT under the supervision of Professors Regina Barzilay and Tommi Jaakkola, alongside researchers from both MIT and Recursion.
