A MAGIBU-BASED MODEL FOR PEDIATRIC AND JUVENILE CNS TUMORS: AN IN-HOUSE EPIGENETIC DECISION-SUPPORT FRAMEWORK COMPARED WITH ONLINE DNA METHYLATION CLASSIFIERS

ABSTRACT: DNA methylation profiling is widely used for CNS tumor classification. We developed MAGIBU, an in-house framework that integrates DNA methylation data via UMAP projection onto a reference CNS tumor landscape and ranks tumor classes based on epigenetic proximity. Rather than providing pre-defined diagnoses, MAGIBU offers quantitative information to support integrative assessment. We evaluated MAGIBU in eight morphologically challenging pediatric CNS tumors with unresolved histopathology at institutional and central review. Reference diagnoses were defined based on the seven cases showing concordant results between the Heidelberg CNS Tumor Methylation Classifier and Methylscape Analysis, two widely used classifiers in clinical practice, serving as a consensus reference. Epigenomic Digital Pathology (EpiDiP), a complementary dimensionality-reduction-based approach, was also applied. EpiDiP was discordant in five of seven cases, positioning low-grade tumors among higher-grade categories. In contrast, MAGIBU was discordant in two of seven cases: one major discordance, a diffuse leptomeningeal glioneuronal tumor positioned near anaplastic pilocytic astrocytomas, and one minor imprecision, an infratentorial FGFR1-altered pilocytic astrocytoma clustering with other infratentorial pilocytic astrocytomas without recognition of its molecular subtype. Both cases in which MAGIBU classifications diverged from reference diagnoses involved rare tumor entities that are minimally represented or entirely absent in existing reference dataset. In the eighth case, unclassifiable by Heidelberg CNS Tumor Methylation Classifier and assigned by Methylscape Analysis to ganglioglioma, EpiDiP indicated similarity to high-grade glioma, whereas MAGIBU positioned the tumor near non-neoplastic tissue. Overall, MAGIBU aligned with consensus diagnoses from established methylation classifiers in the majority of cases and provides a quantitative, differential framework that supports the integration of clinical features, tumor location, patient age, and additional molecular analyses in diagnostically ambiguous cases.

ORGANISM(S): Homo sapiens

PROVIDER: GSE319846 | GEO | 2026/09/09

REPOSITORIES: GEO