AI is designing proteins that could help treat cancer

Using AI, scientists have built a “GPS” for cancer-fighting immune cells.

It’s a molecular navigation system that helped the cells lock onto cancer, kind of like Google Maps guiding travelers to a new address, researchers report July 24 in Science. At its core, the system relies on tiny proteins custom-designed by artificial intelligence.

The approach, a form of immunotherapy, is “very much a proof-of-concept,” says Timothy Jenkins, a medical biotechnologist at the Technical University of Denmark in Lyngby. But his team’s end goal is to develop new therapies doctors can use to treat cancer ­— perhaps even personalized for individual patients.

“It’s an exciting advance,” says Stanley Riddell, an immunotherapy researcher at Fred Hutch Cancer Center in Seattle. Though the work is still in the early stages, it showcases the power of AI models for synthetic protein design, he says. Such models are “likely to generate a whole new class of therapeutics for a variety of diseases that will go beyond cancer.”

Earlier this year, in fact, Jenkins and his colleagues reported a similar AI advance in a different field of medicine: AI-designed proteins that could lead to improved antivenoms for snakebites. Now, the researchers have directed their AI systems toward a new destination: cancer.

The team was looking for ways to rev up the cancer-seeking potential of immune cells called T cells. T cells can fight cancer on their own but sometimes have trouble recognizing the enemy. Jenkins’ team genetically engineered T cells to carry tiny custom proteins on their surface. Those proteins act as the GPS, guiding T cells to their cancer target. The work riffs off other immunotherapy techniques, like CAR T-cell therapy and TCR therapy, which also try to boost immune cells’ anti-cancer prowess.

To design the custom proteins, the researchers relied on a trio of AI tools. First, the team inputted the structure of the cancer target into a generative AI model called RFdiffusion. That model had been trained on known protein structures and their amino acid sequences, the strings of building blocks that fold up into individual proteins. RFdiffusion proposed protein shapes that fit the target like a key fits a lock. A second AI model suggested strings of amino acids that, when folded into 3-D structures, would likely form the proposed shapes.

Jenkins and his colleagues then blasted through tens of thousands of protein designs and, with the help of a third AI model that checked all that work, narrowed the designs down to 44 options that they tested in the lab. One appeared to be a winner. In lab experiments, human T cells engineered to have the AI-designed protein on their surface could rapidly kill melanoma cells and prevent the cancer from growing.

The team’s work is based on the computational protein design and structure prediction technologies that led to the 2024 Nobel Prize in chemistry.

It takes as little as a day or two to come up with promising designs, Jenkins says, and just a few weeks to test them in the lab. That’s faster than current methods, which can involve hunting through people’s cells to pick out proteins, known as T cell receptors, naturally able to lock on to specific cancer targets. “It’s extremely laborious,” says Christopher Klebanoff, a medical oncologist and researcher at Memorial Sloan Kettering Cancer Center in New York City. The process can take months and even then “you can end up with nothing or a very, very small number of therapeutic candidates.”

Klebanoff thinks the new work is an important step, but he’s curious how the AI-designed proteins will work inside the body. Before the researchers take their approach to clinical trials in humans, they’ll need to do many more tests in the lab and in animals, which could take years, says Kristoffer Haurum Johansen, a synthetic immunologist also at the Technical University of Denmark. But for now, he says, his team’s work “means that there’s a potential new tool in the toolbox that we can use to design and develop new therapies.”