More than a half-dozen public and private sector organizations are teaming up to develop biology datasets for artificial intelligence models.
The consortium, which announced the collaboration today, is committing $1.8 billion to the project.
Studying how a cell responds to a new therapy usually requires scientists to set up a physical test environment. That can take years of work and millions of dollars. A faster, more cost-efficient approach would be to test therapies using simulated cells. A simulation doesn’t require expensive lab equipment and can theoretically be spun up in a few hours.
Ultra-detailed, highly reliable simulations that can substitute lab equipment aren’t yet available. The reason is that creating such virtual cells would require a new generation of biology-optimized AI models. Training such models would only be possible with bigger biology datasets than those presently available to scientists.
The research initiative announced today aims to produce the necessary datasets. The biggest contributor to the project is the U.S. Department of Energy, which will commit more than $500 million over five years.
Some of the funds are earmarked for microscopy equipment. The hardware will be used to collect cell measurements for AI training datasets.
The Energy Department will prioritize three imaging methods: neutron scattering, X-ray and cryo-electron microscopy. The first two methods identify the materials that make up a biological sample. Neutron scattering equipment spots light elements and their isotopes, while X-ray machines detect heavier elements. Cryo-electron microscopy, the third imaging method on the list, maps out the internal structure of cells.
The second biggest contributor to the project is Biohub, a research nonprofit backed by Mark Zuckerberg. It’s committing $400 million to the initiative. Three quarters of the funds will go toward in-house research endeavors, while the rest will support external projects.
Like the Energy Department, Biohub will use cryo-electron microscopy to collect cell data. The nonprofit plans to use a specific variant of the technology called cryo-ET. It works by freezing a biological sample and exposing it to an electron beam. When the electrons come into contact with the sample, they change their course and thereby provide insight into its internal structure.
Biohub’s work will also extend to other areas. It plans to collect data about not only the inner workings of cells but also how different cells interact with one another. According to the group, the plan is to create images that contain millions to billions of cells.
The U.S. National Institutes of Health, or NIH, is also participating. It will provide access to biology datasets developed “through more than $500 million in prior federal investment aligned to this initiative.” Biohub and NIH will team up to organize the data into a form that can be more easily processed by AI models.
Google DeepMind, fellow Alphabet Inc. subsidiary Isomorphic Labs and Meta Platforms Inc. are together contributing $300 million. Nvidia Corp., meanwhile, plans to support aspects of the project that will use accelerators and specialized software to run research workloads. The tech giants will be joined by a half-dozen nonprofit research groups including the Allen Institute.
“An accurate predictive model of biology could dramatically accelerate scientific discovery by enabling scientists to perform experiments digitally,” said Biohub head of science Alex Rives. “The insights that come from this could unlock a far greater understanding of disease and open up completely new paths for cures.”