AI CAR T-cell therapy bests approved treatment in mouse study
Researchers say AI may someday help design better cancer treatments
Written by |
Artificial intelligence (AI) may help design better CAR T-cell therapies to treat myeloma and other cancers, a study showed.
An AI-designed CAR T-cell therapy controlled myeloma more effectively than an approved treatment of the same class in a mouse model of the disease.
“We basically saw complete control of the cancer using our AI-designed binder, whereas the existing clinical product could not limit tumor growth,” Caleb Lareau, PhD, co-author of the study and a computational biologist at Memorial Sloan Kettering Cancer Center, said in a news story from the center.
Researchers say the study could serve as a first step towards eventually using AI to help design personalized cancer therapies.
“You can imagine a patient coming in with a rare blood cancer or sarcoma or even a rare pediatric cancer, where nothing off the shelf is going to work — and we’ll be able to use AI to make a therapy that will recognize their cancer within a matter of weeks,” Lareau said.
The study, “Sequence and structural determinants of efficacious de novo chimaeric antigen receptors,” was published in Nature Biomedical Engineering.
Binder is key to CAR T-cell treatment
CAR T-cells are a type of immunotherapy that harnesses the cancer-killing abilities of T-cells. In this type of therapy, a patient’s T-cells are equipped with a human-made protein, called a chimeric antigen receptor (CAR), designed to bind to a specific protein expressed on cancer cells. When the CAR binds to its target, it triggers the T-cell to attack and kill the cancer cells.
One of the most important considerations in making an effective CAR is the binder — the part of the receptor that physically sticks to its target. Ideally, a binder will stick very tightly to the cancer protein, but won’t stick to other proteins found on healthy cells.
Binders have traditionally been created using antibody fragments. Since antibodies can bind specific targets with high specificity, they are useful for this purpose — but identifying suitable binders requires extensive trial and error, which slows development. Antibodies are also relatively large molecules, which can cause logistical complications when designing CARs.
The advent of AI has led to an explosion of research using AI-designed proteins. Rather than starting with an existing protein like an antibody, these tools use computer programs to build brand-new proteins from scratch.
Here, researchers applied a series of AI tools to identify binders that could be used in CARs targeting BCMA, a protein expressed by myeloma cells that is targeted by multiple existing CAR T-cell therapies, as well as CD19 and CD22, two proteins expressed in other types of blood cancers.
The team first used an AI tool to generate about a million potential binders. Then, a second AI tool was used to screen the binders, narrowing the poll to a couple of hundred candidates, a number more manageable for experimental testing.
“You could think of it like generating a million answers to a question in ChatGPT and then asking a different AI model, like Claude, to evaluate them,” Lareau said.
The researchers then tested their binders in CAR T-cells. For these, the T-cells were equipped with CARs that were identical except for the binder component. The results against BCMA were the most notable: In a standard myeloma mouse model, the AI-designed CAR was just as effective at controlling cancer as the approved therapy, and in a second model with higher cancer cell ratios, the AI-designed therapy outperformed the approved treatment.
These results “demonstrate that [AI-]designed CARs can be engineered to possess favorable properties for improved tumor control in high tumor burden preclinical models,” the researchers wrote.
‘On the right track’ to developing new treatments
In addition to using AI to design new CARs, the researchers developed an AI tool called CARPNN (CAR Protein Neural Network), which identifies specific molecular properties that make a given binder more or less effective.
“While the overall structures of these candidate proteins will be very similar, we found that tweaks in the amino acid sequences affect how they perform,” Lareau said. Amino acids are the building blocks of proteins.
Through CARPNN and other analyses, the researchers gained key insights into how binder molecular traits affect CAR activity. For example, when binders had strong positive charges, CAR T-cells were more likely to be prematurely activated in the absence of their target. This kind of premature activation can make this type of therapy less effective and also trigger side effects. The researchers also showed that they could use CARPNN to help minimize off-target binding.
“This ability to actually say, ‘This is why a binder does or doesn’t work,’ is a level of understanding that hasn’t been achievable in the CAR field to date,” Lareau said.
The scientists stressed that more work is needed before this approach can be translated to human studies, but they hope AI will help improve the next generation of CAR T-cell therapies, as well as other anticancer treatments.
“A major goal of our group is to use these generative models to develop new cancer drugs, conduct clinical trials, and see them used to help patients,” Lareau said. “And the remarkable BCMA results, I think, show we’re on the right track.”
Leave a comment
Fill in the required fields to post. Your email address will not be published.