Multiview classification of pediatric medulloblastoma refines prognostic stratification and enables imaging-based risk prediction

Despite advances in molecular subclassification, significant heterogeneity within the four primary medulloblastoma subgroups complicates therapeutic design. High-risk SHH and Group 3/4 cases, particularly those with metastases, frequently fail conventional therapy, highlighting the urgent need for molecularly-based risk refinement and improved disease estimation. By integrating five distinct molecular layers—non-synonymous variants, copy number variations (CNVs), DNA-methylation, gene expression, and alternative splicing—we identified 14 multi-omic clusters (MOCs) across 152 primary tumors from the Children’s Brain Tumor Network (CBTN). Within Group 3, we identified two prognostically distinct MOCs (p = 0.00024); favorable outcomes were linked to high-fidelity genomic maintenance and proteostasis, while poor prognosis was characterized by Notch signaling and metabolic reprogramming (IDH2/3B). Similarly, Group 4 MOCs showed divergent outcomes (p = 0.039) driven by either RTK-mediated proliferation or epigenetic stability. Germline analysis revealed that chromatin-modifying variants (SIRT2, PRDM2) associated with favorable clusters, while invasion-associated variants (PTPRT, EPHA1) linked to tumor aggression. In an effort to explore potential avenues for clinical translation, we modeled MRI and histopathology features as surrogates for these molecularly-defined risks. This integrated multi-omic approach refines our understanding of medulloblastoma outcomes, suggesting that integrated omics coupled with clinical imaging can significantly enhance precision medicine and prognostication.