Meta Unveils AI Model and Chemistry Dataset to Accelerate Scientific Discovery
MENLON PARK, CA — May 27, 2025 Meta announced the release of "Open Molecules 2025," a comprehensive chemistry dataset, alongside a new AI model named UMA (Universal Frontier Model for Atoms). This initiative aims to expedite scientific research in fields such as drug development and materials science.
The Open Molecules 2025 dataset encompasses 100 million quantum mechanical simulations, generated over 6 billion compute hours. These simulations, based on Density Functional Theory (DFT), model atomic and molecular interactions across small molecules, biomolecules, metal complexes, and electrolytes. Notably, the dataset includes systems with up to 350 atoms, surpassing the typical 20–30 atom range in previous datasets, offering a more detailed representation of real-world atomic interactions.
Leveraging this dataset, Meta's Fundamental AI Research (FAIR) team developed UMA, an AI model capable of performing DFT-level calculations up to 10,000 times faster. This acceleration enables researchers to simulate thousands of molecular interactions in minutes, potentially revolutionizing the pace of scientific discovery.
Meta has open-sourced the Open Molecules 2025 dataset, encouraging global collaboration and innovation. The company also ensured the exclusion of data that could facilitate the development of biological or radiological weapons, addressing security concerns
This release underscores Meta's commitment to advancing AI and scientific research, providing tools that could lead to significant breakthroughs in various scientific domains.