# A $1.8 billion biology coalition tackles AI’s missing experimental data

Agent contract: 1.2.0

New experiments, measurement technology and shared standards could help researchers predict how cells respond to interventions and choose better laboratory experiments.

Canonical: https://brightaifuture.com/discoveries/biohub-virtual-biology-coalition
Format: discovery
Source publication: 2026-10-07
Bright publication: 2026-10-07
Substantive update: None recorded
Evidence and review: Emerging; confidence: unassessed; approved; ai-assisted. Primary institutional announcements and independent Reuters and Axios reporting establish the coalition’s commitments. Research capabilities, public-access terms and downstream health outcomes require further evidence.

## Editorial image

Rows of pipette tips on an automated liquid-handling robot at the National Cancer Institute’s Cancer Genomics Research Laboratory.: https://brightaifuture.com/media/content/92fcb89dc0ceb7128020d0c372eacf927f4a9a87ad510860b1c2d043adf1307a.jpg

Credit: Automated liquid-handling robots process DNA at the National Cancer Institute’s Cancer Genomics Research Laboratory. Archival photograph, published in 2020. National Cancer Institute / Unsplash.. License: Unsplash License · https://unsplash.com/license.

Source: https://unsplash.com/photos/gray-laboratory-machine-to8o0bqOA6Q

contextual; not the new coalition’s already-built infrastructure, a completed universal cell model or a treatment outcome.

## The story

Biohub, the U.S. Department of Energy, the National Institutes of Health, Google DeepMind, Isomorphic Labs and Meta announced an expanded effort on October 7 to build the experimental data that predictive models of biology need. The coalition puts a reported $1.8 billion of funding, data, computing and measurement technology behind the Virtual Biology Initiative. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/)

The potential benefit is practical: help scientists choose better experiments. A sufficiently reliable model could forecast how a cell responds to a change, allowing researchers to explore possibilities digitally and concentrate laboratory work on the most promising questions. NIH describes possible uses including prioritizing drug targets and interventions for laboratory and clinical testing. [S2](https://www.nih.gov/news-events/news-releases/nih-joins-effort-build-si-ready-data-predictive-models-human-biology)

Today’s announcement establishes a coordinated commitment. It does not report a newly completed universal cell model, a clinical result or a treatment ready for patients. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/)

## Why the missing data matters

A cell atlas can show which cells exist and which genes are active. A useful predictive model must also learn what changes when researchers intervene. That requires measurements across many cell types, conditions and interventions, rather than a larger collection of similar snapshots. NIH identifies both the breadth of these measurements and the limits of today’s instruments as challenges. [S2](https://www.nih.gov/news-events/news-releases/nih-joins-effort-build-si-ready-data-predictive-models-human-biology)

Researchers call a deliberate change a perturbation. Comparing a changed cell with suitable controls helps separate a response to an intervention from background variation. For a model intended to guide experiments, learning those responses is more useful than simply producing a plausible-looking description of a cell. This is the scientific rationale for focusing on controlled experiments and their measured outcomes. [S8](https://arcinstitute.org/news/virtual-cell-challenge-2026)

Biohub’s planned technologies include cryo-electron tomography, which reconstructs structures inside cells in three dimensions, advanced microscopy and tools for engineering biological systems. Its April announcement also describes combining molecular, spatial and dynamic measurements. The ambition spans several scales of biology, rather than relying on one assay. [S6](https://biohub.org/news/virtual-biology-initiative/)

## The work between a measurement and a usable dataset

The coalition plans shared standards, common identifiers and a single point of access so that results from different groups can work together. DOE’s contribution draws on national-laboratory capabilities including exascale computing, structural measurement facilities and autonomous laboratories. The release names the Joint Genome Institute and Environmental Molecular Sciences Laboratory among the available assets. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/)

Standardization sounds administrative, but it affects scientific usefulness. Researchers need to know which intervention was applied, what was measured and whether results from different instruments or laboratories are comparable. Quality checks, controls and clear descriptions of experimental conditions should make it harder for models to learn laboratory-specific artifacts. These are criteria for judging the planned resource, not completed capabilities established by this announcement.

## What the $1.8 billion includes

The total combines different kinds of contributions. DOE plans more than $500 million over five years. Google DeepMind, Isomorphic Labs and Meta are collectively investing $300 million. Biohub’s $500 million anchor commitment was announced in April. NIH is coordinating relevant repositories and datasets developed through more than $500 million in earlier federal investment. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/) [S4](https://www.investing.com/news/stock-market-news/usgovernment-google-join-zuckerbergbacked-biohub-in-18-billion-push-for-ai-biology-data-4936795)

Readers should therefore avoid treating the headline as $1.8 billion of newly announced cash. Existing public research resources are part of the package. The companies’ individual contributions are not broken out in the release. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/)

Biohub’s April allocation was $400 million for its own technology and data-generation work and $100 million for external research. Those investments remain the initiative’s foundation. [S6](https://biohub.org/news/virtual-biology-initiative/)

## “Open” comes with a timing qualification

Public access is a central promise, but there is an important distinction. Reuters reports that commercially funded datasets will initially be available to the funding companies before public release. Axios reports that this exclusive period is one year. Reuters also reports that the parallel government-funded work will not carry that commercial restriction. [S4](https://www.investing.com/news/stock-market-news/usgovernment-google-join-zuckerbergbacked-biohub-in-18-billion-push-for-ai-biology-data-4936795) [S5](https://www.axios.com/2026/10/07/zuckerberg-biohub-ai-biology-data)

The announcements do not supply a comprehensive dataset license, a release-by-release access schedule or a finalized governance charter. A promise of eventual public availability leaves questions about permitted reuse, distribution and how researchers will access the material. Those details deserve scrutiny as the infrastructure takes shape. [S1](https://biohub.org/news/virtual-biology-initiative-expansion/) [S3](https://www.isomorphiclabs.com/articles/isomorphic-labs-joins-the-virtual-biology-initiative)

Human data also require care. NIH’s existing guidance covers informed consent, participant privacy and when controlled access adds protection. The coalition’s announcement does not establish that every underlying human dataset can be freely redistributed. Responsible openness must preserve the conditions under which participants contributed their data. [S9](https://www.grants.nih.gov/policy-and-compliance/policy-topics/sharing-policies/dms/privacy)

## Better models need better tests

There is genuine progress to build on, alongside unresolved measurement problems. A peer-reviewed Nature Biotechnology paper published October 1 examined 14 perturbation datasets and 18 evaluation metrics. It found that deep-learning models could outperform uninformative baselines when metrics were properly calibrated. The authors also warned that different metrics capture different aspects of performance. [S7](https://www.nature.com/articles/s41587-026-03307-w)

Crucially, that study did not address prediction in unseen cell types, donors or conditions. A good result on one task cannot establish a universal ability to simulate biology. The distinction matters for models expected to work beyond familiar training examples. [S7](https://www.nature.com/articles/s41587-026-03307-w)

Arc Institute’s current Virtual Cell Challenge targets a harder question: predicting intervention responses in cellular contexts a model has never seen perturbed. Experimental responses are withheld for evaluation. That offers a useful example of the kind of public, held-out testing the broader field needs, although its results would not automatically validate this coalition’s future models. [S8](https://arcinstitute.org/news/virtual-cell-challenge-2026)

## Who could benefit, and what to watch

Academic laboratories and smaller research teams could benefit if comparable data become accessible without each group repeating expensive measurements. Companies get an earlier research opportunity through the reported embargo. Public benefit will depend on the resource’s usefulness after release, including whether researchers with smaller computing budgets can actually work with it.

Reuters reports that Biohub’s Alex Rives expects a first dataset in about a year and accurate predictive models within five years. These are expectations from the initiative’s scientific leader. The report does not establish a general public-download date for every dataset. [S4](https://www.investing.com/news/stock-market-news/usgovernment-google-join-zuckerbergbacked-biohub-in-18-billion-push-for-ai-biology-data-4936795)

The next meaningful milestones are tangible: released datasets with clear terms; reproducible measurements and documentation; predictions tested on genuinely new biological conditions; and laboratory confirmation of useful discoveries. Faster drug development remains a hoped-for downstream consequence. The coalition’s immediate contribution is to organize and fund the difficult experimental work that could make such progress possible.

## Provenance and history

{
  "dates": {
    "captureDate": "2026-10-07",
    "eventDate": "2026-10-07",
    "lastReviewedDate": "2026-10-07",
    "publicationDate": "2026-10-07"
  },
  "provenance": {
    "origin": "editorial",
    "externalId": "https://biohub.org/news/virtual-biology-initiative-expansion/"
  },
  "revisions": [
    {
      "id": "revision:biohub-virtual-biology-coalition-first-publication",
      "recordedAt": "2026-10-07",
      "sourceIds": [
        "source:biohub-virtual-biology-coalition-s1",
        "source:biohub-virtual-biology-coalition-s2",
        "source:biohub-virtual-biology-coalition-s3",
        "source:biohub-virtual-biology-coalition-s4",
        "source:biohub-virtual-biology-coalition-s5",
        "source:biohub-virtual-biology-coalition-s6",
        "source:biohub-virtual-biology-coalition-s7",
        "source:biohub-virtual-biology-coalition-s8",
        "source:biohub-virtual-biology-coalition-s9",
        "source:biohub-virtual-biology-coalition-s10",
        "source:biohub-virtual-biology-coalition-s11"
      ],
      "summary": "A $1.8 billion biology coalition tackles AI’s missing experimental data"
    }
  ],
  "corrections": []
}

## Original sources

- [S1 · Biohub: October 7 expansion announcement · 2026-10-07](https://biohub.org/news/virtual-biology-initiative-expansion/)
- [S2 · NIH: NIH joins effort to build SI-ready data for predictive models of human biology · 2026-10-07](https://www.nih.gov/news-events/news-releases/nih-joins-effort-build-si-ready-data-predictive-models-human-biology)
- [S3 · Isomorphic Labs: founding-member announcement and official release PDF · 2026-10-07](https://www.isomorphiclabs.com/articles/isomorphic-labs-joins-the-virtual-biology-initiative)
- [S4 · Reuters / Krystal Hu, syndicated by Investing.com · 2026-10-07](https://www.investing.com/news/stock-market-news/usgovernment-google-join-zuckerbergbacked-biohub-in-18-billion-push-for-ai-biology-data-4936795)
- [S5 · Axios / Ina Fried: Zuckerberg teams with Google, U.S. in push to map cells · 2026-10-07](https://www.axios.com/2026/10/07/zuckerberg-biohub-ai-biology-data)
- [S6 · Biohub: Virtual Biology Initiative launch · 2026-04-29](https://biohub.org/news/virtual-biology-initiative/)
- [S7 · Miller et al., Nature Biotechnology: Deep learning perturbation models can outperform baselines on calibrated metrics · 2026-10-01](https://www.nature.com/articles/s41587-026-03307-w)
- [S8 · Arc Institute: 2026 Virtual Cell Challenge · 2026-08-20](https://arcinstitute.org/news/virtual-cell-challenge-2026)
- [S9 · NIH: Data Management and Sharing privacy guidance · 2025-08-05](https://www.grants.nih.gov/policy-and-compliance/policy-topics/sharing-policies/dms/privacy)
- [S10 · Biohub-issued release, Newswise DOE Science News Source copy; primary release syndication · 2026-10-07](https://www.newswise.com/doescience/international-cross-sector-collaboration-commits-nearly-2-billion-to-build-foundational-data-for-ai-models-to-predict-and-treat-disease/?article_id=856083)
- [S11 · Biohub-issued release, PRNewswire; primary release distribution · 2026-10-07](https://www.prnewswire.com/news-releases/international-cross-sector-collaboration-commits-nearly-2-billion-to-build-foundational-data-for-ai-models-to-predict-and-treat-disease-302901172.html)

## Continue exploring

- [Seeing the shape of life.](https://brightaifuture.com/discoveries/alphafold)
- [Seeing molecules together.](https://brightaifuture.com/discoveries/af3)
- [What National Compute’s research preview offers](https://brightaifuture.com/discoveries/national-compute-research-preview-access)
- [Health & biology](https://brightaifuture.com/worlds/health)
- [What becomes possible when we can see and shape life’s machinery?](https://brightaifuture.com/threads/biology)