AI-Designed B Cell–Reactive Neoantigens Open a New Chapter for Personalized Cancer Vaccines

Recently published in Science Advances, researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed an AI-based approach that broadens how neoantigens are identified for personalized cancer vaccines. Instead of focusing only on T cell responses, which is the standard approach in most vaccine pipelines, the team’s model also predicts whether neoantigens can activate B cells and trigger antibody production. 

Using machine learning trained on both structural and sequence features of mutant proteins, KAIST’s researchers managed to identify B cell–reactive neoantigens at scale. In preclinical studies, cancer vaccines incorporating these antigens produced stronger antitumor immune responses, and analyses of existing clinical trial data suggested that accounting for B cell reactivity is associated with improved vaccine efficacy. 

Building on these findings, KAIST has partnered with Neogenlogic, a Seoul-based biotech company dedicated to developing AI-powered personalized cancer therapy, to advance the technology toward preclinical development, with plans to pursue an FDA Investigational New Drug (IND) application and enter clinical trials in the coming years.

Understanding Neoantigens and Their Central Role in Cancer Vaccines

Neoantigens are protein fragments derived from cancer cells that serve as a unique marker for distinguishing cancer from other cells. Given the fact that these peptides are absent from normal tissues, they offer highly specific targets for immunotherapy with reduced risk of off-tumor toxicity. Over the last decade, neoantigen-based vaccines have become a central pillar of personalized cancer immunotherapy, powered by advances in tumor sequencing, epitope prediction and vaccine delivery platforms (peptide, mRNA, dendritic cell). Despite technical progress, clinical responses have been variable, and the development of effective neoantigen selection remains a critical bottleneck for achieving consistent, durable benefit.

Most current neoantigen discovery pipelines focus on identifying peptides that bind to major histocompatibility complex (MHC) molecules and activate T cell receptors. This approach is based on the well-established model in which cytotoxic CD8⁺ T cells recognize neoantigens presented on tumor cells and directly kill those cells. However, this T cell–centered strategy largely overlooks how B cells recognize antigens. Unlike T cell epitopes, B cell epitopes often depend on the three-dimensional structure of a protein rather than a simple linear sequence. Because of this structural complexity, conventional sequence-based prediction methods struggle to identify neoantigens capable of triggering antibody responses. As a result, vaccine designs that focus only on T cell immunogenicity may fail to capture antigens that could engage B cells and activate additional, complementary antitumor immune mechanisms. 

How KAIST’s AI Platform Overcomes Existing Limitations

KAIST’s AI platform goes beyond traditional T cell–focused approaches by modeling how mutant proteins interact with B cell receptors (BCRs). Instead of relying only on MHC-binding predictions or T cell receptor motifs, their deep learning model evaluates whether a mutated peptide is likely to form the three-dimensional structures needed for B cell recognition and antibody production. Importantly, the platform does not treat B cell reactivity as a separate or optional feature. It combines predicted T cell and B cell responses into a single scoring system to prioritize neoantigens, reflecting the idea that a coordinated immune response involving both antibodies and T cells can produce stronger and more durable antitumor effects.

The current study developed the B cell predictor using an exceptionally large training dataset. The KAIST researchers have analyzed over 437,000 peptides for IgG binding and examined more than 370 millions B cell receptor (BCR) sequences to identify patterns associated with antibody recognition. They validated the new AI model using single-cell BCR sequencing, which cracks the genetic code of the antibody-producing receptors on individual B cells, and then applied the multiomics framework to large genomic and clinical cohorts, including samples from the Cancer Genome Atlas (TCGA) and checkpoint-blockade response datasets (2074 patients), to explore associations at the population level. By combining molecular training data, animal experiments, and retrospective clinical analyses, the study provides strong evidence that B cell–reactive neoantigens can meaningfully enhance antitumor immune responses.

The Impact of B Cell–Reactive Neoantigens

In mouse vaccination experiments, neoantigens selected for B cell reactivity produced effects beyond what T cell–focused vaccines achieved. These vaccines triggered expansion of BCR clones and, in several experimental groups, led to faster tumor regression compared with vaccines targeting only T cell responses. 

In clinical analyses, retrospective analysis across multiple vaccine trials and checkpoint inhibitor studies indicated that integrating predicted B cell reactivity correlated with stronger immune signatures. Also, a meta-analysis of 11 personalized vaccine trials involving 1739 neoantigens suggested that incorporating B cell neoepitopes may improve vaccination efficacy. Overall, these findings suggest that B cell–reactive neoantigens can significantly enhance antitumor responses in both laboratory experiments and patient data, providing a compelling rationale for their inclusion in personalized cancer vaccines.

Translational Pathway, Commercialization Plans and Implications for Industry

KAIST has partnered with Neogenlogic to commercialize the AI platform. According to Prof. Jung Kyoon Choi, one of the co-first authors of the current study and CEO (Research and Development) of Neogenlogic, “We are conducting pre-clinical development of a personalized cancer vaccine platform and are preparing to submit an IND application to the FDA with the goal of entering first-in-human clinical trials in 2027. We will enhance the scientific completeness of cancer vaccine development based on our proprietary AI technology and push forward the transition to the clinical stage step-by-step.” 

For biotech companies and investors in the field of cancer vaccine development, KAIST’s approach represents a shift from focusing solely on T cell targets to evaluating multiple aspects of the immune response. Implementation of this strategy will require robust structural modeling, standardized assays to measure B cell activity, and carefully designed clinical trials that track both cellular and antibody responses. Regulatory approval will also depend on clear frameworks to monitor immune outcomes, including antibody production, memory B cell formation, and their impact on patient benefits.

Author

Richard Chau