MethPed

MethPed classifies pediatric brain tumors into biologically defined subtypes using DNA methylation profiles to support research and diagnostic interpretation.

Key Features:

• Multiclass Random Forest Algorithm: Implements a multiclass random forest classifier tailored to analyze DNA methylation data.
• Data Source: Trained on methylation data from Illumina Infinium HumanMethylation450 BeadChip arrays obtained from the Gene Expression Omnibus (GEO) comprising 472 samples.
• Validation: Performance was validated using DNA from 18 regional samples.
• Performance Metric: Reports classification performance with a Cohen's kappa statistic of 0.98.
• Application Scope: Predicts subclasses for nine pediatric brain tumor subgroups: Diffuse Intrinsic Pontine Glioma (DIPG), Ependymoma, Embryonal Tumor with Multilayered Rosettes (ETMR), Glioblastoma (GBM), Medulloblastoma group 3 (MB3), Medulloblastoma group 4 (MB4), WNT-activated Medulloblastoma (MB_WNT), Sonic Hedgehog-activated Medulloblastoma (MB_SHH), and Pilocytic Astrocytoma (PiloAstro).

Scientific Applications:

• Diagnostic Classification: Provides subclass predictions to confirm histologic diagnoses of pediatric brain tumors.
• Differential Diagnosis: Aids in suggesting alternative or mixed-morphology diagnoses for challenging cases.
• Molecular Research and Therapeutics: Supports molecular characterization of tumor subgroups and development of targeted therapeutic hypotheses.

Methodology:

Uses Illumina Infinium HumanMethylation450 BeadChip array data (GEO, n=472) with validation on 18 regional DNA samples and a multiclass random forest classifier evaluated using Cohen's kappa (0.98).