TEIP: A Compact, Open-Source Framework for Predicting Tumor Epitope Immunogenicity in Glioblastoma Cancer Using Deep Learning and Multi-Modal Biological Features

This work introduces a modular, open-source computational pipeline for glioblastoma (GBM) vaccine design that integrates omics-based OIP5 target discovery with a deep learning framework for epitope immunogenicity prediction. Building upon conventional affinity-based predictors such as NetMHCpan, our Tumor Epitope Immunogenicity Pipeline (TEIP) incorporates biological, structural, and transcriptomic context to predict tumor-specific T-cell responses. Using curated immunogenic and non-immunogenic peptides from IEDB and GBM datasets, TEIP employs dual bidirectional LSTM encoders to represent peptide and HLA sequences, concatenated with auxiliary molecular features including proteasomal cleavage, TAP transport likelihood, gene expression, and mutation frequency. These features are fused into a probabilistic model that outputs peptide-specific immunogenicity scores and confidence estimates. Benchmark results across multiple HLA alleles demonstrate that TEIP outperforms NetMHCpan and MARIA, achieving an ROC-AUC of 0.89 and PR-AUC of 0.35. Feature importance analysis confirms that tumor-specific expression and peptide-MHC binding dominate predictive accuracy, validating the biological realism of the model. By combining mechanistic features with data-driven representation learning, TEIP enables fine-grained prioritization of neoantigens with translational potential. In proof-of-concept GBM studies, TEIP recovered known cancer-testis antigens such as OIP5, demonstrating its ability to identify clinically relevant vaccine targets. The pipeline is implemented using entirely open-access data and software to promote reproducibility and scalability. Collectively, TEIP provides a unified framework that connects multi-omics tumor characterization with deep learning-based antigen modeling, establishing a foundation for precision immunotherapy development in glioblastoma and beyond.