Mark Zuckerberg’s Biohub is partnering with Google DeepMind, Isomorphic Labs, Meta, U.S. federal agencies and a collection of scientific organizations to generate large, standardized biological datasets and develop AI models that can predict how cells behave. Biohub describes the initiative as an effort to create a “universal virtual cell.”
Why this matters
The aim is to let scientists run many more hypotheses and simulated experiments virtually so that costly, time-consuming laboratory work is reserved for the experiments most likely to provide useful empirical insight. While AI has made progress at understanding biological components such as proteins, modeling an entire living cell is far more complex and currently limited by the lack of sufficient empirical data.
How the project will proceed
The first phase will assemble a broad map of cellular biology by collecting diverse types of measurements on cells and their responses to perturbations. In later stages the team envisions models that could analyze an individual’s disease, infer its molecular causes and predict the best interventions.
Funding and data access
The effort combines roughly $1.8 billion in funding, data, computing and measurement technology. The U.S. Department of Energy plans to invest more than $500 million over five years in biological measurement, modeling and computing. The National Institutes of Health is contributing datasets and resources stemming from more than $500 million in previous federal investments. Google DeepMind, Isomorphic Labs and Meta are jointly contributing about $300 million, and Biohub previously committed $500 million to the initiative.
Commercial partners will have one year of exclusive access to the data they develop before those data are released publicly. Alex Rives, head of science at Biohub, said the temporary exclusivity is intended to incentivize private companies to invest while ensuring the resulting data rapidly become an open scientific resource.
Scientific significance and challenges
Rives noted that recent AI advances have combined better algorithms, greater computing power and huge amounts of data. Biology adds an extra difficulty: much of the data AI needs do not yet exist and must be painstakingly measured from the physical world. He characterizes the necessary models as “empirical AI” — models that learn from biological evidence and can accurately predict real-world physical outcomes. Rives called the effort the start of a new scientific paradigm linking computation and the digital domain to the physical realities of life.
If successful, such models could help researchers study fundamental questions like aging and regeneration, and medical questions such as molecular mechanisms behind Alzheimer’s disease.
Timeline and expectations
Rives said researchers should not have to wait long to assess whether the approach is working: within about a year of assembling the first large-scale dataset, teams should be able to train models, measure capabilities and determine which additional kinds of biological data improve performance. He pointed to AI’s progress in protein biology as an example of how increased data and compute can drive capabilities.
Summary
The Biohub-led initiative brings together tech companies and federal research institutions to standardize and expand biological data and to develop AI-driven cellular models. Backed by substantial funding and a one-year data embargo for commercial contributors, the project aims to accelerate biological discovery by enabling more experiments to be evaluated virtually before committing to laboratory validation.
