PATH-01. The NCI/Bethesda DNA Methylation Classifier: An Important Diagnostic Tool for CNS Tumors

DNA methylation-based classification has become essential for accurately diagnosing CNS tumor subtypes by providing a reproducible and objective tool in addition to existing diagnostic modalities. Heidelberg classifier is highly accurate and valuable, we aimed to develop an alternative classifier to: (1) provide a ‘second look’ at methylation results alongside the Heidelberg classifier, and (2) enable the inclusion of recently described tumor classes. The NCI/Bethesda classifier version 3.0 was developed using a reference set of 19,530 samples where classes (n = 128) are aligned with tumor diagnostic categories, including those in the WHO classification. On the external validation set (n = 4,475), the NCI/Bethesda classifier achieved a higher match rate (47%, 2,096/4475 samples) compared to the Heidelberg classifier (37%; 1,658/4,475 samples) using classifier-specific confidence thresholds (0.85: DKFZ; 0.95: Bethesda classifier). High-confidence matches were achieved for 1,145/4,475 samples by both classifiers, 951 samples were uniquely matched by the NCI/Bethesda classifier, compared to 513 uniquely matched by the Heidelberg classifier. To examine the clinical utility of the NCI/Bethesda classifier, 796 samples with low-confidence ‘suggestive’ (0.5-0.84) scores in the Heidelberg classifier were examined by our classifier, showing high confidence matches in 296/796 cases (37%), a number of which could be confirmed using orthogonal methods, suggesting that a subset of cases would benefit from our classifier. We also find that our classifier is more robust for high tumor heterogeneity and low tumor purity in subset of tumor types such as GBMs, O-IDH and PXA. The NCI/Bethesda classifier provides a needed and complementary tool for methylation-based classification of CNS tumors, that is partially orthogonal to the Heidelberg classifier. Of note, the NCI/Bethesda classifier includes several classes absent from the Heidelberg classifier, such as Intracranial mesenchymal tumor and Desmoplastic myxoid tumor, among others. The NCI/Bethesda classifier is freely available for use (https://methylscape.ccr.cancer.gov).