AI-designed cell-surface tags boost antigen display for vaccines and CAR-T therapies

Scientists report today in Nature Biotechnology that a deep learning suite called DeepSCan can discover and design potent cell-surface display (CSD) elements — short protein modules that anchor antigens or receptors to the outer membrane of cells. The work, led by corresponding author Sidi Chen at Yale School of Medicine, with first author Zhenhao Fang, maps the sequence rules that make these display tags work, then uses them to generate new tags that outperform the strongest naturally occurring ones.

The team first measured surface expression across more than 570 chimeric antigens and assigned cell-surface translocation strength labels to roughly 310 CSD elements. They then trained three generations of deep learning models, which guided the design of about 3,700 generative CSDs. Experimental testing of around 120 of these identified 7 generative CSDs that matched or exceeded the surface display potency of the best natural CSDs. The top designs were validated across multiple cell lines, and one was shown to functionally present the cancer target BCMA on Raji cells, enabling BCMA-directed CAR-T cells to kill those cells in a dose-dependent manner.

The findings matter because many emerging mRNA vaccines and cell-based immunotherapies rely on displaying antigens or signaling receptors on the cell surface to trigger immune responses. Until now, CSD elements were chosen mostly from known proteins without a clear understanding of what makes them potent. DeepSCan provides a systematic, AI-driven route to engineer these modules, potentially improving vaccine immunogenicity and the effectiveness of engineered cellular therapies.

The study is a preclinical, cell-based proof of concept; it does not include animal or human data. The authors also note that model predictions for sequences far from the training distribution should be used with caution. Still, the results show that machine learning can not only classify natural display elements, but also create new ones with enhanced function — opening the door to more rational design of synthetic membrane proteins and antigen-presenting platforms.

Reference: Fang, Z., Saskin, J., Lee, SH. et al. Discovery and design of potent cell surface display elements. Nature Biotechnology (2026). https://doi.org/10.1038/s41587-026-03144-x