Google DeepMind, Meta, and Alphabet's AI drug discovery company Isomorphic Labs are collectively contributing $300 million to Biohub's Virtual Biology Initiative. The nonprofit research organization, founded by Mark Zuckerberg and Priscilla Chan, aims to build a "virtual cell" using artificial intelligence to predict cellular behavior and responses. This initiative is part of a larger $1.8 billion effort to create comprehensive AI datasets for biological research.
The Virtual Biology Initiative seeks to generate and standardize biological data at scale, addressing a key bottleneck in AI development for biology. Unlike language models that can learn from existing text, AI in biology often requires new experimental data to be created. Biohub plans to use this funding to develop technologies for measuring biological processes with greater detail, including cryo-electron tomography and advanced microscopy. The goal is to allow researchers to digitally simulate experiments, potentially accelerating the discovery of new treatments for diseases.
The total $1.8 billion for the Virtual Biology Initiative includes Biohub's initial $500 million commitment, over $500 million in planned spending from the U.S. Department of Energy, and existing datasets from the National Institutes of Health (NIH) developed with over $500 million in prior federal investment. The Department of Energy will invest in lab measurement, modeling, and computation, while the NIH will coordinate and standardize existing data repositories for AI training. Nvidia is providing computing infrastructure and software, and other scientific partners include the Broad Institute and the Allen Institute.
Biohub, established in 2016 with initial funding from the Chan Zuckerberg Initiative, has a mission to cure and prevent all disease. The organization previously launched projects such as a cell atlas and has been developing AI models for protein biology. Commercial partners involved in the initiative will receive a one-year exclusive access period to the data they help generate before it is made publicly available. Biohub expects to release its first large dataset within approximately one year and aims for accurate predictive models within five years.