Biotechnology · global
Filling the Experimental Data Gap for “Virtual Cells”: Biohub Expands $1.8 Billion Initiative with the U.S. Government and Tech Industry
For AI to predict how cells respond to drugs, researchers must first measure how life changes. This cross-sector collaboration will integrate new experiments, computing resources and existing data, but open sharing will follow different timelines, and its effectiveness in treating disease has yet to be validated.
If researchers could first predict on a computer how cells respond to drugs or environmental changes, they might be able to determine earlier which treatment ideas are worth taking into the laboratory. The challenge is that models need large amounts of comparable data from real experiments. On October 7, Biohub announced an expanded collaboration with the U.S. Department of Energy (DOE), the National Institutes of Health (NIH) and technology industry partners, bringing together funding, data, computing and measurement technologies valued at a total of $1.8 billion to lay the groundwork for this kind of “virtual cell” research.
That total is not all newly committed cash. Biohub had already pledged $500 million when it launched its “Virtual Biology Initiative” in April this year; DOE will contribute more than $500 million over five years, while Meta, Google DeepMind and Isomorphic Labs will jointly contribute $300 million. NIH’s contribution is to coordinate the inclusion of datasets, databases and knowledge resources developed through more than $500 million in past federal investment, with Biohub helping organize them into formats suitable for model training.
The initiative aims to fill gaps in data on how cells change under different conditions. Beyond describing a cell’s state at a single moment, researchers also need to know how molecular activity, cellular structures and interactions between cells change after an intervention. If models can reliably capture these responses, they could help explore disease mechanisms and screen treatment hypotheses; this is also why the partners want to expand the range of cell types and experimental conditions they measure.
DOE, NIH and Biohub have signed a memorandum of understanding. According to DOE’s announcement, the department will bring together supercomputers capable of performing a quintillion calculations per second, imaging and structural measurement facilities at national laboratories, and automated experimental systems, including the OPAL program under the Genesis Mission. These resources will support data generation and validation, with the data then used for model training and cell simulation, connecting experiments with computing.
Another hurdle in collaboration across institutions is making the data truly usable together. Biohub plans to establish shared standards, common identifiers and a unified access point to reduce barriers to connecting different experiments and databases. Isomorphic Labs also separately confirmed that it has become a founding member of this international initiative and will help build data resources for the research community. For models, whether data can be compared and experimental conditions traced is as important to the credibility of predictions as the volume of data.
However, “open” does not mean that all data will be released publicly at the same time. According to Reuters, Biohub Chief Scientist Alex Rives said that data supported by commercial funders would be subject to an embargo before public release, allowing the relevant teams to use it first before it becomes a public research resource; government-funded work would have no such restriction. The report did not specify the length of the embargo, so when the research community can access each batch of data still depends on subsequent arrangements.
Rives told Reuters that the partners hope to have the first datasets ready in about a year and to concentrate the overall effort within five years; these are planned targets. The announcement provided no model prediction accuracy figures, independent validation results or evidence of clinical efficacy. Moving from cellular responses to treatment effects in patients will still require step-by-step confirmation through experiments and clinical research. The first question this commitment must answer is whether it can establish sufficiently comprehensive, reusable biological data to provide a reliable basis for predictions in disease research.