DISP-13. DNA methylation-based ethnicity prediction enables analysis of racial disparities in molecularly classified CNS tumors

Abstract BACKGROUND Central nervous system (CNS) tumors exhibit significant racial disparities, with recent studies showing distinct distributions among molecularly classified CNS tumors. African Americans are overrepresented in pituitary adenomas (35%) while underrepresented in IDH wild-type glioblastoma (3% vs. 90% in Caucasians), suggesting ethnic factors contribute to tumor development. However, current large-scale databases including SEER, CBTRUS lack ethnicity annotation with current WHO molecular classification. This prevents analysis of ethnic disparities across the >100 molecularly distinct CNS tumor entities now recognized, representing a critical gap in understanding tumor predisposition and developing precision medicine approaches. METHODS We developed a machine learning classifier to predict ethnic backgrounds of CNS tumor samples using DNA methylation arrays, the same data used for molecular classification. The classifier was trained on an ethnically heterogeneous public dataset comprising healthy blood and brain tumor samples, and will be calibrated on a separate brain tumor dataset to ensure accurate performance for CNS tumor samples. RESULTS Our classifier demonstrated excellent validation performance with log loss of 0.188 and mean AUROC of 0.995 across ethnic groups. Individual performance showed consistently high AUROC (African American = 0.995, Caucasian = 0.994, East Asian/Asian American = 0.998, Hispanic/Latino/South American = 0.993). Mean Brier scores were low at 0.023 overall, indicating good calibration across ethnic groups (African American = 0.012, Caucasian = 0.038, East Asian/Asian American = 0.013, Hispanic/Latino/South American = 0.028). Following calibration, the classifier will be applied to over 150,000 CNS tumor samples to systematically analyze ethnic disparities across CNS tumor subtypes. CONCLUSIONS We developed an accurate ethnicity prediction tool that will enable large-scale analysis of ethnic disparities in molecularly classified CNS tumors. This addresses gaps in current epidemiological analyses and will provide insights into how ethnicity influences tumor biology and molecular classification patterns, potentially informing precision medicine approaches for diverse patient populations.