AT A GLANCE:
A team led by Marinka Zitnik , associate professor of biomedical informatics in the Blavatnik Institute at Harvard Medical School, has received a grant of nearly $7 million from Cancer Grand Challenges to create artificial intelligence (AI) “co-scientists” that help cancer researchers analyze data, generate hypotheses, and suggest new experiments, speeding new knowledge and therapies.
The project, known as AURORA, includes an additional $4 million to three international teams across Spain, the United Kingdom, and the United States, which will evaluate the effectiveness of the “co-scientists” in their ongoing studies of next-generation CAR T-cell therapies, cancer initiation, and tumor-immune interaction.
“By evaluating agentic AI co-scientists within selected Cancer Grand Challenges research teams, we hope to understand whether they can increase the efficiency and effectiveness of researchers, achieving advances in oncological research far beyond what humans can achieve alone,” said Zitnik, who also serves as associate faculty at the Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University, associate member at the Broad Institute of MIT and Harvard , and affiliated faculty at the Harvard Data Science Initiative .
“Because of its complexity and unmet medical need, cancer research provides a unique test bed to demonstrate this new mode of doing science through human-AI collaborations,” she added. “If we’re successful, this system could generate insights that lead to the development of life-saving treatments.”
Cancer Grand Challenges is a global research initiative that identifies the toughest problems in cancer research today and supports a global community of world-class, interdisciplinary research teams in coming together to take them on. It was founded in 2020 by the two largest funders of cancer research in the world — Cancer Research UK and the National Cancer Institute in the United States.
The AURORA project is funded by Cancer Research UK through Cancer Grand Challenges. It marks the first Cancer Grand Challenges project on AI and the first that spans multiple Cancer Grand Challenges-funded teams.
An AI member of the research team
AI models have grown increasingly sophisticated over the past several years and hold promise in many areas of biomedical research. Although AI has demonstrated capacity to reproduce known biological findings, Zitnik explained, systems that generate new hypotheses, test them in the lab, and learn from the results are still at an early stage. This need is especially critical for oncology research — a field marked by an incredible wealth and complexity of data.
To fill this gap, Zitnik is leading a team of scientists based in her lab to develop an agentic AI system that works alongside researchers through the full cycle of discovery. The system reasons across evidence from many sources, uses a large collection of linked scientific tools to propose testable hypotheses, and refines its reasoning as experimental results come in. It builds upon other AI tools developed at Harvard, including Zitnik’s ToolUniverse .
The new agentic system will serve as a conversational partner to human researchers. But unlike chatbots, the new AI collaborator will have the capacity to access and integrate all data available in existing scientific literature; databases; other biological AI tools such as AlphaFold, protein language models, and omics-based tumor models; and real-time experimental findings.
After calling on this vast repository of knowledge, the AI co-scientist can suggest new hypotheses and develop ideas for experiments to confirm or refute them, prioritizing time, budget, or whatever factors are important to the human investigators. It can then analyze the data from those experiments to suggest new scientific directions.
In the AURORA project, humans remain responsible for all scientific judgement and decision-making.
This process has the potential to significantly speed the pace and reduce the cost of research by eliminating many dead ends, Zitnik said. Rather than testing an idea through brute force — for example, individually silencing hundreds or thousands of genes to find one responsible for a specific disease feature — scientists will be able to pursue the most likely candidates from the start, shrinking the pool of possibilities and the time needed to find the answer.
Improvement over time
Evaluation is embedded throughout the system, Zitnik explained. Besides her own team’s AURORA grant, three research groups already leading other Cancer Grand Challenges projects, known as Cancer Grand Challenges Pioneer Teams, will receive approximately $1.3 million each in funding to add the AI co-scientist to their teams. These projects are led by scientists from the following teams:
Through frequent interactions with their “team member” powered by agentic AI, these groups will provide constant feedback that will test and improve the system over time — for example, changing the data it prioritizes to suggest ideas or preventing it from suggesting impractical experiments.
“A key challenge for AI in science is understanding how it can best support researchers working on real-world problems,” said Zitnik. “By testing the system against real scientific questions and experimental evidence, we hope to build a trusted, rigorously evaluated tool that helps researchers accelerate discovery.”
Transparent reasoning, tested in the lab
Rather than its guidance being a “black box” like some AI systems, with no way to trace the logic behind its suggestions, AURORA’s AI co-scientists will provide reasoning traces that allow users to understand the rationale behind every output. This feature is critical both to foster helpful feedback from the researchers using it and to encourage their trust in the system, Zitnik said.
As Cancer Grand Challenges Pioneer Teams test the system’s hypotheses in the lab, the AURORA team will compare its reasoning with the experimental results to measure how often AURORA’s AI co-scientists lead to findings that hold up. Over time, Zitnik hopes this approach will establish AI as a partner across biomedical research, well beyond cancer.
The AURORA team launched the project on Nov. 5 at Harvard, introducing the Pioneer Teams and hosting a panel of speakers from academia and frontier AI labs on what an AI would have to get right to join a cancer research team.