An international team led by St. Jude’s Children’s Research Hospital and Heidelberg’s Hopp Children’s Cancer Center reported M‑PACT, a methylation‑based predictive algorithm that classifies pediatric CNS tumors from trace ctDNA in cerebrospinal fluid. Published in Nature Cancer, the ensemble deep‑learning model achieved 92% classification in benchmarking and 88% in a validation cohort. M‑PACT uses EM‑Seq to preserve low‑input methylomes and imputes missing CpG data to estimate tumor fraction and subtype — a potential tool to guide treatment strategies and clinical trial enrollment when tissue biopsies are infeasible.