Investigating the Underlying Molecular Patterns in Aggressive Brain Tumor ETMR to Develop a Robust Diagnostic Prediction Model

Rare, aggressive pediatric brain tumor Embryonal tumor with multilayered rosettes (ETMR) has been minimally researched and is frequently misdiagnosed, reducing chances of survival. While ETMR is characterized by a general amplification in the C19MC gene cluster, this isn't a unique molecular identifier. This study aims to differentiate ETMR from other brain tumors through specific gene amplification patterns within C19MC and development of a robust diagnostic model. Various publicly available methylation patient data was compiled and fine-tuning methods to reduce model memorization and optimize the limited available data. The most amplified genes in the C19MC cluster and genes with high feature-importance from the model were found to consist of oncogenes controlling ETMR indicators. Many tumor suppressor genes were also present, potentially due to their deactivation from hypermethylation. These amplifications provide more specific hallmarks and potential targets to enhance therapeutic treatment. Additionally, testing an unseen dataset on the diagnostic model after training with fivefold cross validation revealed a 0.932 Area Under the Curve (AUC), 0.916 specificity, and 0.833 sensitivity to differentiate between children with or without ETMR from epigenetic factors. This novel machine learning application can assist physicians with prompt identification and therefore reduced mortality rates of ETMR.