C-BRAIN

C-BRAIN (the Consortium for Biomedical Research and AI in Neurodegeneration, styled "C-brAIn" in early coverage) is a consortium of academic researchers, pharmaceutical companies, philanthropic organizations and data and literature repositories that builds open AI tools for Alzheimer's disease and neurodegeneration research. It was founded and is directed by neurologist Randall J. Bateman of Washington University School of Medicine in St. Louis (WashU Medicine), which led its formation.^[1][2][7] The consortium's stated goal is an "AI Biomedical Research Scientist" that works alongside human researchers, motivated by the fact that more than 99% of Alzheimer's drug candidates fail in clinical trials.^[1] Its released software includes InsightEngine, a three-agent research platform, and OpenScientist, an open-source agentic "co-scientist" developed with Lawrence Berkeley National Laboratory.^[3][4] In October 2026 NVIDIA announced a collaboration with the consortium on compute and model optimization.^[5]

History

Bateman first pitched the idea, then called the Artificial Intelligence Biomedical Research Scientist Initiative, to about 60 scientists at a meeting alongside the AD/PD conference in Vienna in April 2025. Alzforum reported that stakeholders from nearly 60 organizations had shown interest, including Microsoft and Google, and that Bateman estimated about $10 million to start the project and $100 million in total for the platform's computing needs.^[6] Bateman formally announced the consortium's launch at a 30 July 2025 meeting held at the Alzheimer's Association International Conference (AAIC) in Toronto, where it began a three-year plan; by then it had funding from five philanthropic groups and 10 major contributing members.^[7] NVIDIA's October 2026 blog post gives the launch as November 2025, a date that does not match the consortium's own news page or the Alzforum report.^[5][8]

Organization and members

C-BRAIN's leadership includes Bateman as director, Eric Landsness (associate director, a WashU Medicine assistant professor of neurology), Ade Ojewole as chief technology officer, Anna M. Santacruz as operations program leader, and Adith Jagadish Boloor as technical lead for its Azure and NAIRR deployment.^[2][1] Bateman holds the Charles F. and Joanne Knight Distinguished Professor of Neurology chair at WashU Medicine.^[1]

The July 2026 WashU announcement described C-BRAIN as a 17-member consortium and listed 15 major contributing members and two key members; the consortium's team page lists the same 17 organizations.^[1][2] NVIDIA's October 2026 post described it as having 18 member institutions.^[5]

WashU described the arrangement as a pre-competitive space for pharmaceutical members to work on biological targets and mechanisms before drug development, while philanthropic backers fund tools that remain non-commercial and openly available to approved researchers.^[1] The consortium uses a federated design in which members keep control of their own data, so proprietary and unpublished pharmaceutical data can inform the tools without being exposed or transferred, with a "scientist-in-the-loop" approach that keeps human researchers involved at each stage.^[1] NVIDIA described this as a hybrid of centralized and federated approaches, with some data deposited directly into secure systems.^[5]

Software tools

The 13 July 2026 release covered three interrelated tools, which WashU said were developed by Boloor, Ojewole and Landsness, built in part with NAIRR Pilot resources.^[1] All three are free to biomedical researchers in neurodegeneration who register with C-BRAIN for approval, and a public demonstration is available on the consortium's website.^[1]

These three functions had been planned from the start: Alzforum's 2025 reports described a first year of "minimum viable products" consisting of a literature synthesis system combining large language models, knowledge graphs and retrieval-augmented generation, the integration of unpublished "dark data", and a "reviewer 3" service.^[6][7] C-BRAIN uses "dark data" to mean unpublished findings, negative results and governed proprietary data from member institutions.^[5]

InsightEngine

On the consortium's website, InsightEngine is the platform that combines three agents sharing a common knowledge graph and provenance layer, covering literature and data synthesis, hypothesis testing against experimental data, and review of manuscripts and proposals.^[4]

• The literature review and data synthesis agent ingests Alzheimer's literature, including PubMed articles and consortium datasets, extracts entities and relationships into a persistent knowledge graph, and links each statement in its answers to a source publication. Domain experts can flag or amend extracted facts.^[4]
• The data analysis agent takes a hypothesis, writes a test plan (datasets, replication checks, variables and analysis steps), harmonizes inputs into a common schema, and runs statistical or code-based analyses in a sandbox, reporting likely failure points, confidence levels and reasons for discrepancies between studies.^[4]
• The critical reasoning agent parses manuscripts, proposals and reports, including figures and tables, and evaluates them for significance, innovation, approach, rigor and ethics from editorial, methodological and domain-expert perspectives.^[4]

C-BRAIN says the platform supports switchable underlying models and is containerized for deployment on AWS, Google Cloud or Azure.^[4] Its source code is split across four public GitHub repositories (web interface, backend API, AI service and user service), each under the Apache 2.0 license.^[4][12] NVIDIA said InsightEngine was built with 387 Labs and that it checks claims against evidence to catch hallucinations before researchers commit to experiments.^[5]

OpenScientist

OpenScientist is an open-source, domain-agnostic discovery agent led by Justin Reese of Lawrence Berkeley National Laboratory and released under the Apache 2.0 license.^[3][13] Given data files and a research question, it runs an iterative loop (10 iterations by default) of generating hypotheses, testing them with sandboxed Python code and searching PubMed, then writes a final report.^[3][13] According to its preprint, it is built on Anthropic's Claude Code with a public collection of Agent Skills, and Claude Sonnet 4.5 was used for every run reported in the paper.^[3] The repository exposes its tools through the Model Context Protocol, can reach models through Lawrence Berkeley's CBORG service, Google Vertex AI, AWS Bedrock or Azure AI Foundry, tracks per-project cost, and has optional structural biology support through Phenix, including AlphaFold confidence analysis.^[13] A hosted instance runs at openscientist.io.^[3]

The preprint positions OpenScientist against other AI scientist systems, including Google's AI co-scientist, Sakana's AI Scientist, Edison Scientific's Kosmos and Biomni, arguing that most such platforms are closed-source and therefore hard to verify independently.^[3]

OpenScientist evaluation

The preprint, "OpenScientist: evaluating an open agentic AI co-scientist to accelerate biomedical discovery" (Roberts et al., posted on medRxiv on 18 March 2026, not peer reviewed), had domain experts evaluate the system in four case studies.^[3]

The authors report that OpenScientist completed in minutes analyses that would otherwise take weeks to months of human time, and that a typical run in these case studies cost less than $10.^[3] They also documented failures: in 10 early runs of the SEABIRD case, the system treated empty CSV cells as zeros (misclassifying participants with missing PET data as amyloid negative), used a literature-derived Centiloid cutoff instead of the prespecified threshold, counted duplicate records as independent observations, and inconsistently reported sample sizes and confidence intervals. These were resolved through prompt refinement and data curation.^[3] NVIDIA's blog cites these case studies and projects that, across C-BRAIN's reach of more than 200 research groups, the capability could free more than one million research hours a year at scale; that figure is NVIDIA's projection, not a measured result.^[5]

Computing and the NVIDIA collaboration

On 27 October 2026 NVIDIA published a blog post describing its work with C-BRAIN, and the consortium announced that it is collaborating with NVIDIA to optimize open-access AI models of Alzheimer's disease.^[5][8] According to NVIDIA, C-BRAIN runs 64 NVIDIA data center GPUs on premises alongside cloud nodes on Microsoft Azure, with further capacity, including NVIDIA Blackwell systems, planned. The two are working to bring the NVIDIA BioNeMo platform and the MONAI medical imaging framework into C-BRAIN's research pipelines.^[5] Ojewole said that optimization would be needed as the consortium pushes the limits of current spatial omics machine learning architectures.^[5]

Funding

Alzforum reported in August 2025 that the consortium had pledges of $6 million toward a projected $10 million budget for its first-phase tools, and that the second phase could cost as much as $50 million, in part because of the computing needed to train on petabytes of data.^[7] In May 2026 WashU Medicine said it had been awarded nearly $800,000 by the NAIRR Pilot program and Microsoft to develop the consortium's "AI Scientist" system, with an initial focus on Alzheimer's disease.^[11] The Alzheimer's Drug Discovery Foundation, Bristol Myers Squibb and Gates Ventures are among the major contributing members.^[1]

Goals

The 2025 plan set targets of expanding the first-year tools to at least 50 labs by July 2026, testing an integrated AI assistant in more than 100 labs by July 2027, and having an AI biomedical scientist ready to support labs across the Alzheimer's research community by July 2028.^[7] When Bateman asked Alzheimer's scientists at the April 2025 meeting whether a three-year timeline was achievable, most said no; he gave it "a fair 50 percent chance, maybe higher."^[6] According to NVIDIA, C-BRAIN also plans a multimodal biological foundation model trained on neuroimaging, multi-omics, fluid biomarker and longitudinal clinical data, and aims to identify new causal mechanisms of brain aging and Alzheimer's disease by its third year.^[5]

See also

• AI for Science
• AI in drug discovery
• Agentic AI
• FutureHouse
• NVIDIA BioNeMo

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Shawn Ballard and Jessica Church, "Global consortium launches AI tools to accelerate Alzheimer's research, treatments," WashU Medicine / The Source, 13 July 2026. source.washu.edu/...alzheimers-research-treatments
2. ^1 ^2 ^3"Our C-BRAIN Team," C-BRAIN. c-brain.org/team
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14Kaleigh F. Roberts et al. (corresponding authors Randall J. Bateman and Justin T. Reese), "OpenScientist: evaluating an open agentic AI co-scientist to accelerate biomedical discovery," medRxiv, posted 18 March 2026. doi.org/...2026.03.15.26348338
4. ^1 ^2 ^3 ^4 ^5 ^6 ^7"InsightEngine AI," C-BRAIN. c-brain.org/...insightengine
5. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Anthony Costa, "How AI Is Changing the Pace of Discovery for Millions Living With Dementia," NVIDIA Blog, 27 October 2026. blogs.nvidia.com/...ource-ai-tools-cbrain-dementia
6. ^1 ^2 ^3 ^4Jessica Shugart, "New Partnership Forms to Harness AI for Alzheimer's Research," Alzforum, 11 April 2025. alzforum.org/...rms-harness-ai-alzheimers-research
7. ^1 ^2 ^3 ^4 ^5 ^6Jessica Shugart, "AI Biomedical Scientist in Its Sights, C-brAIn Gets Going," Alzforum, 25 August 2025. alzforum.org/...tist-its-sights-c-brain-gets-going
8. ^1 ^2"News & Events," C-BRAIN. c-brain.org/news-events
9. ^Kaleigh F. Roberts et al., "Towards an AI biomedical scientist: Accelerating discoveries in neurodegenerative disease," The Journal of Prevention of Alzheimer's Disease 13 (2026): 100398. doi.org/...j.tjpad.2025.100398
10. ^"C-BRAIN at AD/PD 2026 Copenhagen," C-BRAIN. c-brain.org/adpd-workshop
11. ^1 ^2Katie Cammarata, "WashU Medicine researchers to build AI tools to boost Alzheimer's research," WashU / The Record, 4 May 2026. source.washu.edu/...s-to-boost-alzheimers-research
12. ^"C-brAIn-InsightEngine-ai," GitHub. github.com/...C-brAIn-InsightEngine-ai
13. ^1 ^2 ^3"OpenScientist: Scientific Hypothesis Agent for Novel Discovery," GitHub repository openscientist-io/openscientist. github.com/...openscientist