NeoBridge: Biophysics-informed neoantigen prioritization and clonal architecture profiling in pancreatic cancer

Accurate identification of immunogenic neoantigens remains a bottleneck in personalized cancer vaccine development, particularly in low-mutation-burden tumors such as pancreatic ductal adenocarcinoma (PDAC). We developed NeoBridge, a hybrid ensemble integrating gradient boosting and a deep neural network across four biologically grounded feature dimensions: HLA binding and processing, differential agretopicity, tumor neoantigen abundance, and TCR recognition potential. Trained on 11,188 experimentally validated neoantigens, NeoBridge achieved an area under the receiver operating characteristic curve (AUC-ROC) of 0.904 on held-out data, with cross-cancer generalization (AUC: 0.73-0.82), including 0.82 in pancreatic cancer, and outperformed established tools. Differential agretopicity emerged as the dominant predictive feature. In 118 PDAC patients with resolved phylogenies and complete follow up, subclonal neoantigen load was associated with worse progression-free survival (p < 0.001) and overall survival (p = 0.038). These findings are hypothesis generating and require prospective immunopeptidomic and functional validation.