EPCO-46. EPIGENETIC STEMNESS PREDICTION ACROSS CENTRAL NERVOUS SYSTEM TUMORS MOLECULAR SUBTYPES

Abstract Tumor progression and therapeutic resistance in cancer have been strongly associated with stemness. Rather than focus on a discrete subpopulation for stemness tumor maintenance, we might explore as continuous variable through machine learning algorithms. We chose One Class Logistic Regression algorithm to define an epigenetic signature assess brain tumors using induced Neural Stem Cell (iNSC) public DNA methylation data. In order keep features most related iNSC, perform Wilcoxon test between 9 iNSC 128 non-tumor tissue (methylation difference |0.3|, FDR < 0.01) mapped the resultant genome regions. Our model revealed positive correlation (r2 = 0.86 p 0.001) pluripotent from Malta et al. (2018) applied TCGA gliomas. More remarkable, stratified IDHwt glioma survival by median of both GLASS cohorts (TCGA: 0.001, GLASS: 0.086; Likelihood ratio test). Having shown its potential, next other Central Nervous System (CNS) studied Capper (2016). The distinct methylation-based subtypes, such as: i) ependymoma-RELA presented very lowest among all ependymal subtypes; ii) low lymphoma high plasmacytoma; iii) pituitary-ACTH pituitary-STH-DNS-B. Interestingly, gliomas IDHmutant Additionally, putative new entities identified Capper, our high-grade-neuroepithelial-BCOR embryonal infantile-hemispheric-glioma, high-grade-neuroepithelial-MN1, anaplastic-pilocytic-astrocytoma showing differences each “Other glioma” group. results indicate prognosis prediction recapitulate CNS subgroups, which suggest surrogate strategy perspective.