Computational Identification of Clonal Neoantigens in IDH-Wildtype Glioblastoma Through Integrated Genomic Filtering and HLA Prediction

Background Glioblastoma (GBM) remains highly resistant to immunotherapy despite advances in neoantigen-based approaches. While whole-exome sequencing (WES) comprehensively identifies somatic mutations, resulting candidate pools are enriched for subclonal and passenger variants with limited immunogenic potential. We developed a data-driven pipeline integrating mutation frequency, clonality (VAF > 0.4), and expression (TPM > 10) to systematically prioritize clonal driver neoantigens in IDH-wildtype GBM, reducing analytical complexity while enriching for therapeutically relevant targets. Methods We analyzed 127 IDH-wildtype GBM patients from The Cancer Genome Atlas (TCGA), applying multi-criteria filtering based on mutation frequency, variant allele frequency (VAF > 0.4 for clonality), and gene expression (TPM > 10). Candidate genes were prioritized for neoantigen prediction using NetMHCpan-4.2 (HLA class I) and NetMHCIIpan-4.3 (HLA class II) with population-based HLA panels covering ~70% (class I) and ~60% (class II) of global populations. Strong binders were defined as peptides with IC50 < 500 nM and percentile rank < 2%. Results Our computational pipeline identified three genes ( TP53, PTEN, EGFR ) meeting all filtering criteria, representing frequently mutated, clonally enriched (60-66% clonal mutations), and highly expressed loci in GBM. From 181 clonal mutations across these genes, we predicted 186 unique neoantigen peptides (150 class I, 36 class II) generating 468 HLA-peptide binding predictions. These neoantigens covered 87 of 127 patients (68.5%), with TP53 and PTEN contributing most candidates. Class I epitopes dominated (80.6%), consistent with GBM's CD8+ T-cell-centric immune landscape. However, 98% of neoantigens were patient-specific, with minimal sharing across individuals, reinforcing the need for personalized immunotherapy approaches. Conclusions We demonstrate that systematic filtering for clonal, expressed driver mutations provides an efficient alternative to WES-based neoantigen discovery, achieving substantial patient coverage while dramatically reducing analytical complexity. Our generalizable, data-driven approach identifies high-confidence neoantigen candidates suitable for personalized vaccine development and complements emerging machine learning-based immunogenicity prediction tools. The modular, reproducible pipeline and focus on evolutionarily stable targets addresses key limitations of current neoantigen discovery workflows in low-mutation-burden cancers. Future validation through peptide-MHC binding assays and patient-specific HLA typing will be essential to translate these computational predictions into clinical immunotherapies.