← Back to Home

Can AI Choose the Right Next Step in Cancer Research? Harvard’s AURORA Project Moves Into Laboratory Validation

Three cancer research teams will involve AI in generating hypotheses and prioritizing experiments, testing whether this “co-scientist” can deliver useful discoveries in areas ranging from childhood solid tumors to the mechanisms of cancer initiation. Whether this improves research efficiency is precisely the question the project aims to answer.

By SURL BioNews

Cancer research has no shortage of leads; the challenge is deciding which ones warrant the next round of experiments. Faced with vast amounts of genetic, cellular and immune data, researchers must make choices within limited time and funding. The AURORA project, led by Harvard Medical School researcher Marinka Zitnik, will use AI to help generate hypotheses and prioritize experiments, with laboratories then testing whether its recommendations hold up. Launched at Harvard University on November 5, 2026, the project also makes the question of whether AI can improve scientific research itself a subject for testing.

AURORA is funded by Cancer Research UK through Cancer Grand Challenges. The charity described this announcement as the next phase of the £8 million project, with the three participating teams receiving a combined additional $4 million in support. The system is built on ToolUniverse, developed by Zitnik’s team, which enables AI to access scientific tools, databases and computing resources to integrate evidence and propose testable ideas, while also considering the time, budget and feasibility of experiments.

The NexTGen team will focus on CAR T-cell therapies for childhood solid tumors, which involve engineering immune cells to recognize specific targets. According to an announcement from University College London, AI will analyze biological data from cancer cells and healthy cells to help prioritize targets, target combinations and cell engineering strategies, and explore why tumors resist treatment. Researchers hope this will narrow the pool of candidates and concentrate experimental resources on better-supported designs.

PROMINENT takes the question back to before cancer develops: Why do some cells carrying cancer-associated mutations remain normal, while others gradually become malignant? On November 13, the Institute for Research in Biomedicine in Barcelona said the team had received an additional $1.31 million in funding, with Núria López-Bigas and Stanford University’s Emma Lundberg jointly leading the work. AI will help organize the literature and large biological datasets, propose mechanistic hypotheses about how mutations alter cells and tissues, and incorporate factors such as age and sex. Selected ideas will then be tested using approaches including genetic screens and cell imaging.

The third team, MATCHMAKERS, focuses on immune recognition: How do T cells recognize cancer cells? The many possible combinations of tumor antigens and T-cell receptors would require substantial resources to test individually. The team will develop interacting AI co-scientists to help select combinations of tumor markers and immune cells worth studying, which researchers will validate in the laboratory, building evidence to understand how the immune system targets tumors.

At the heart of all three research directions is the continuous feedback of experimental results into the system. Researchers will define questions, review hypotheses and select experiments; AI will then adjust subsequent recommendations based on new evidence. Harvard Medical School said the system will also provide records of the reasoning behind its recommendations for researchers to examine. Evaluation will cover which recommendations are supported, which fail and how performance changes as new data are added. Responsibility for scientific judgment and decision-making will remain with people.

The announcements so far describe development and evaluation plans. They have not yet provided quantitative results showing that AI improves experimental success rates, shortens research time or reduces costs, nor have they detailed the full design of controlled comparisons or the scale of the datasets. Whether recommendations withstand experimental testing and whether overall research efficiency improves remain separate questions requiring evidence. AURORA’s value going forward will depend on whether it can consistently help scientists make better-supported choices under real research constraints. Whether patients will benefit will require a longer process of validation.

References

  1. Harvard Medical School
  2. Cancer Grand Challenges
  3. University College London
  4. IRB Barcelona
  5. Cancer Research UK