Alex Rives, the head of science at Biohub, sees a new role for artificial intelligence in biology. “The big challenge in biology is to bridge that gap between compute and the digital world and the real physical world of biology and life. The way to do that is through data,” he said.
The Chan Zuckerberg Biohub is joining the Department of Energy, the National Institutes of Health, Google DeepMind, Isomorphic Labs, Meta and scientific organisations in an effort to create what Biohub calls a “universal virtual cell”. The project aims to produce and standardise biological data that artificial intelligence can use to predict how cells behave.
The partners are pooling $1.8 billion in funding, data, computing and measurement technology. Biohub was founded as a philanthropic effort by Meta chief executive Mark Zuckerberg and his wife, Dr. Priscilla Chan. Zuckerberg and Chan co-founded the organisation, and Chan views the virtual-cell initiative as a community asset that can build on itself over time.
The effort builds on a framework set out in a landmark paper published in Cell in December 2024. Its central challenge is scale: existing datasets cover hundreds of millions of cells, while a predictive model accurate enough for the task will require billions and eventually trillions.
The Department of Energy plans to invest more than $500 million over five years in biological measurement, modelling and computing. The National Institutes of Health will contribute datasets and other resources produced through more than $500 million in earlier federal investment. Google DeepMind, Isomorphic Labs and Meta are collectively putting in another $300 million, while Biohub had already committed $500 million. NVIDIA will provide computing infrastructure and technical expertise.
Among the research bodies that have agreed to work within the effort are the Allen Institute, Broad Institute, Gladstone Institutes, Human Cell Atlas, Human Protein Atlas and Wellcome Sanger Institute. The data will include spatial transcriptomics, a technique that maps molecular activity within intact tissue, alongside screens tracking how cells respond when their environment changes.
The partners want to compress work that might otherwise take decades into five years. An initial dataset is expected in roughly a year. Within a year of receiving that first large-scale dataset, Rives said, researchers should be able to train models, assess their capabilities and establish which further forms of biological data improve them.
Commercial partners will receive one year of exclusive access to the data they develop before it is made public. Functional predictive models are expected within the same five-year period.
The intended result is an AI model through which scientists can run potential experiments virtually, identify the strongest prospects and then decide which are worth the time and cost of laboratory testing.
Comments (0)