When Daniel Baker earned his Ph.D. from the University of Pennsylvania last December, he brought with him a discovery that could fundamentally change how scientists hunt for new cancer therapies. Working under the mentorship of CAR T cell therapy pioneer Carl June, Baker helped develop an AI framework that can identify promising targets for engineered T cell treatments in a matter of weeks—work that traditional methods might take months or even years to accomplish.
The team at Penn's Perelman School of Medicine and Abramson Cancer Center combined the pattern-recognition power of large language models with the deep expertise of human researchers. Their approach, published in the journal Cell, uses multiple frontier AI systems to scan through more than 10,000 potential targets, running independent simulations 1,000 times over to filter out AI errors like hallucinations.
"Discovering a good CAR target is like trying to find a needle in a haystack, except the haystack keeps growing as more sequencing data becomes available," Baker explained. "Human experts excel at going deep, while LLMs are good at looking across a broad range of data. So, we created a framework that combines these strengths."
The team's proof of concept centers on a protein called GPNMB, which their AI-driven analysis identified as the most promising target. When the researchers engineered T cells to recognize and attack cells bearing this marker, the treatment showed robust tumor-killing activity in mouse models of melanoma, leukemia, and colorectal cancer.
What makes this work particularly significant is its accessibility. The framework was built to work with publicly available datasets, a design choice Baker hopes will democratize cancer target discovery. "We hope to make this broadly available beyond teams who have access to clinical samples or major institutions that are able to do their own sequencing," he said.
CAR T cell therapy, which was pioneered at Penn Medicine, has already transformed care for several blood cancers over the past decade. The new AI approach could help extend those gains to solid tumors and other conditions by making the target identification process faster, cheaper, and more widely accessible to researchers around the world.