AI-designed binder outperforms existing CAR-T binders, “completely controlling” cancer

An artificial intelligence (AI)-designed binder was found to control BCMA, a blood cancer target protein, more effectively than the binders used in existing US Food and Drug Administration (FDA)-approved chimeric antigen receptor T-cell (CAR-T) therapies.

According to BioOn on the 20th local time, researchers at Memorial Sloan Kettering Cancer Center (MSK) reported in the international journal Nature Biomedical Engineering that they used generative AI to design a new binder that completely controlled tumor growth in a mouse model.

CAR-T therapy involves genetically modifying a patient’s immune cells so they attack cancer. However, a recent analysis found that only about 4% of patients with advanced or metastatic cancer in the United States are eligible to receive this treatment.

A research team led by Caleb Lareau, PhD, at MSK sought to expand the applicability of CAR-T therapy by improving its binder. A binder is a protein that recognizes and attaches to a target protein on the surface of cancer cells.

Existing FDA-approved CAR-T therapies use an antibody-derived single-chain variable fragment (scFv) as the binder. In contrast, the AI-designed binders were newly created proteins that did not borrow from existing antibodies. They were smaller and more compact, allowing them to attach precisely to specific regions of the target protein.

The research team generated approximately 1 million candidate binders using 2 existing AI tools and narrowed them to 100 to 200 candidates. Dr. Lareau described the process as “creating 1 million answers in ChatGPT and having another AI model, Claude, evaluate them.”

The team then trained its own AI tool, CARPNN, on the experimental results to analyze why certain binders were effective. The researchers found that subtle differences in charge, amino acid composition and binding location had a major effect on performance.

The research team tested the new binders against several targets of FDA-approved therapies. For BCMA, which is found on multiple myeloma cells, the MSK binder significantly inhibited tumor growth compared with the FDA-approved version.

“The AI-designed binder completely controlled the cancer, but the existing clinical product failed to prevent tumor growth,” Dr. Lareau said.

Some experiments also revealed limitations. The AI model failed to produce an effective binder against CD19, a blood cancer target, and binders with a high positive charge or high alanine content showed a problem with premature activation even in the absence of cancer cells.

Based on these successes and failures, the research team established design rules encompassing charge and amino acid composition. “We are turning traditional ‘educated guessing’ into evidence-based design,” Dr. Lareau said.

Dr. Lareau explained that the technology could be used not only to develop CAR-T therapies but also to design binders for other cancer treatments, including precision drug conjugates. “When a patient with a rare blood cancer or sarcoma comes in, we may be able to use AI to create a treatment that recognizes the cancer within a few weeks,” he added.

Postdoctoral researcher Arthur Chao, graduate student Hoiin Chu and research technologist Luofan Li were co-first authors of the study.