From T Cells to B Cells: The Next Generation of Personalized Cancer Vaccines

A research team at the Korea Advanced Institute of Science and Technology (KAIST) has developed a new artificial intelligence model that could change how personalized cancer vaccines are designed. For decades, most cancer vaccine strategies have focused on training the immune system’s T cells to attack tumors quickly and directly, which can shrink tumors but often fails to provide lasting protection. The weakness of this approach is that it is very good at attacking in the short term but relatively poor at creating long-term immune “memory,” which is crucial to prevent cancer from coming back.

At the heart of this new work are neoantigens, which are protein fragments created by mutations that exist only in the tumor and not in normal tissues. These neoantigens act as unique “fingerprints” for each patient’s cancer and are ideal targets for the immune system because they are truly foreign to the body. Personalized cancer vaccines aim to teach the immune system to recognize these neoantigens so it can attack cells carrying them and, ideally, remember them for years.

The adaptive immune system has two major arms: T cells and B cells. T cells are like an elite assault force that can recognize and kill cancer cells displaying certain antigens on their surface, which explains why most current vaccines have been built around activating cytotoxic (killer) T cells. B cells, by contrast, produce antibodies that circulate in the blood and tissues, tag abnormal cells for destruction, and, importantly, generate long-lived memory cells that can respond rapidly if the same antigen appears again in the future. Growing evidence from cancer immunology shows that robust B cell–mediated immunity and the presence of so-called tertiary lymphoid structures in tumors are strongly associated with better responses to immunotherapy and improved survival.
​
The key innovation from KAIST is an AI framework that can predict which neoantigens will not only activate T cells but also elicit a strong B cell response. To build this system, the team trained models on very large datasets that included hundreds of thousands of peptides tested for antibody binding and hundreds of millions of B cell receptor (BCR) sequences. By learning the structural interaction patterns between mutant peptides and BCRs, the AI can estimate which neoantigens are likely to be highly “visible” to B cells and capable of driving strong antibody and memory responses. In practical terms, the workflow begins with sequencing a patient’s tumor, identifying mutations that are unique to the cancer, and then using the AI to score each corresponding neoantigen for both T cell and B cell immunogenicity. Only those with high predicted reactivity on both arms of the immune system are chosen for inclusion in a personalized vaccine.
​
The researchers validated their approach in several ways. In preclinical experiments, vaccines enriched for B cell–reactive neoantigens produced stronger antitumor immune responses and better tumor control than vaccines designed solely around T cell predictions. When B cells were experimentally depleted in animal models, the therapeutic benefit of these vaccines dropped and tumor volumes increased, directly demonstrating that B cells were essential contributors to the antitumor effect. The team also analyzed thousands of tumor samples from The Cancer Genome Atlas and found that mutations predicted to be strongly B cell–reactive were underrepresented in real-world cancers, consistent with the idea that the immune system has already been “editing out” many of the most B cell–visible variants. Finally, by reanalyzing data from multiple personalized cancer vaccine trials, they showed that including B cell–reactive neoantigens correlated with better clinical outcomes than focusing on T cell targets alone.
​
This work, published in Science Advances at the end of 2025, is described as the first AI framework that jointly optimizes neoantigen selection for both B cell and T cell responses in the context of personalized cancer vaccination. The research team has indicated that preparations are underway for an investigational new drug (IND) submission to the U.S. Food and Drug Administration, with the goal of entering early-phase clinical trials around 2027. If successful, this would move the concept from an academic proof-of-principle into a clinical platform where each patient’s vaccine is designed by computationally ranking their tumor’s mutations according to how powerfully they can engage the full adaptive immune system.
​
Although the platform is intended for multiple tumor types, the implications for prostate cancer are notable. Vaccine-based strategies for prostate cancer, including personalized neoantigen approaches, are already being explored in men with locally advanced or oligometastatic disease, where the aim is to harness the immune system to control micrometastatic spread and delay or prevent progression. Integrating B cell–centric design, as pioneered by the KAIST framework, could improve the quality and persistence of vaccine-induced responses in prostate cancer, particularly when combined with existing standards such as androgen deprivation therapy, radiotherapy, or checkpoint inhibitors.
​
Source.