A Deep Learning-Driven Framework Integrating Organoid-Based Functional Validation Identifies Universal Neoantigens from Recurrent Glioma Mutations.

Glioblastoma (GBM) is the most common malignant intracranial tumor in adults, with a median survival of only 16 to 20 months. Neoantigen therapy has shown advantages in the treatment of GBM, as it improves the immunosuppressive microenvironment within the tumor. However, the identification of truly immunogenic neoantigens remains a major challenge. Current computational prediction tools primarily focus on antigen presentation, whereas algorithms that incorporate T-cell immunogenicity features remain limited. Furthermore, standard validation methods, such as enzyme-linked immunospot (ELISpot) assays, lack physiologic relevance and do not fully recapitulate the tumor microenvironment. In this study, we developed a neoantigen prediction algorithm, TCRscore, based on publicly available datasets by integrating human leukocyte antigen binding and T-cell receptor (TCR) recognition features. Twenty-one patient-derived GBM organoid models were established from isocitrate dehydrogenase wild-type tumors to validate the performance of the algorithm. Predicted neoantigens were evaluated using ELISpot assays, flow cytometry, and in vitro killing assays based on organoid-T cell coculture systems. TCRscore outperformed six existing tools in predicting immunogenic neoepitopes. The organoid models retained the key histologic and transcriptomic features of parental tumors and provided an effective platform for functional validation. Coculture assays confirmed that neoantigen-specific T cells could induce targeted killing in GBM organoids. In particular, the analysis identified that the recurrent PIK3R1G376R mutation contributed to a potential shared neoantigen in GBM. Overall, by integrating TCRscore with organoid-based validation, this study provides a high-fidelity, high-quality GBM neoantigen database with significantly enhanced prediction accuracy. SIGNIFICANCE: A clinically impactful framework that integrates a TCR-aware AI algorithm with glioblastoma organoids enables accurate neoantigen prediction and validation, advancing both personalized and population-level immunotherapy strategies.