Enhancing Molecular Classifying Accuracy of Pediatric CNS Tumors: A Dual-Classifier Approach Using DNA Methylation Profiling

Esra Mosa¹Rania Alanany^1,2Shimaa Sherif¹Erdener Ozer¹Sukoluhle Dube¹Aayesha Jabeen¹Ian Pople¹Davide Bedognetti³Ata Maaz¹Ayman Saleh¹william Mifsud¹Wouter R.L. Hendrickx^1,2*Christophe Michel Raynaud^1*

• ¹Sidra Medicine, Doha, Qatar
• ²Hamad Bin Khalifa University College of Health and Life Sciences, Doha, Qatar
• ³Ospedale San Martino, Oristano, Italy

DNA methylation–based classification has improved central nervous system (CNS) tumor diagnostics, but pediatric data on real-world implementation remain limited. We evaluated two DNA methylation–based classifiers— the Heidelberg classifier and the NIH/Bethesda (Methylscape) classifier— in a single-center cohort of pediatric patients. Ninety-six samples from 96 patients (75 CNS tumors, 10 non-CNS tumors, and 11 non-neoplastic CNS lesions) were profiled using Illumina MethylationEPIC arrays (850K/930K). We compared calibrated scores, concordance with integrated histopathological diagnoses, and the impact of technical factors such as tissue preservation, analyzable CpG count, and array version. Methylation classification agreed with integrated histopathology in 88.0% (66/75) of CNS tumors and refined diagnoses in 54.7% (41/75). Both classifiers showed high concordance but occasionally assigned high-confidence labels to non-neoplastic lesions, underscoring the importance of joint pathological review. Fresh frozen versus FFPE tissue, analyzable CpG count, and EPIC v1 versus v2 did not significantly affect classifier performance in our setting. Our findings support the use of methylation classifiers as decision-support tools in pediatric CNS tumor diagnostics, provided that calibrated score thresholds are interpreted in the context of tumor purity, DNA quality, and integrated neuropathology.