- MIT and Recursion (RXRX, Financial) unveil Boltz-2, a revolutionary AI model for rapid and accurate drug discovery.
- Boltz-2 achieves near-FEP accuracy and operates 1000 times faster, transforming virtual screening economics.
- Open-source availability under an MIT license promotes widespread academic and commercial use.
The Massachusetts Institute of Technology (MIT) and Recursion (RXRX) have collaborated to launch Boltz-2, an innovative open-source AI model designed to significantly enhance the drug discovery process. This breakthrough biomolecular foundation model concurrently predicts molecular structure and binding affinity with a precision that approaches industry-standard free energy perturbation (FEP) calculations, but at speeds up to 1000 times faster.
Boltz-2 was trained using Recursion's NVIDIA supercomputer, BioHive-2, and has demonstrated superior performance in the CASP16 affinity challenge. This model employs approximately 5 million binding affinity assay measurements, offering improved affinity prediction and advanced joint modeling capabilities. These features are bolstered by Boltz-steering, enhancing the physical plausibility of simulations.
The model's release under an MIT license marks a significant advancement in the accessibility of AI tools for drug development, allowing both academic and commercial users to leverage this technology. This development represents a major step forward in making large-scale virtual screening more practical and cost-effective, addressing a critical bottleneck in early-stage drug discovery.
For further details, visit the official Boltz-2 page at https://boltz.bio/boltz2.
