Enhancing Nanopore cfDNA diagnostics of pediatric brain tumors with machine learning

Algorithmic methods are redefining diagnostic workflows for brain tumors. Various pipelines for molecular diagnoses based on DNA from solid tumor tissue are used in clinical practice today. While methylation arrays remain the gold standard for methylation profiling, Nanopore sequencing has become a promising method, facilitating overnight diagnostics. Recent works have shown that it can further enable both methylation- and CNV-calling from cerebrospinal fluid (CSF) cell-free DNA (cfDNA), allowing for minimally-invasive tumor diagnoses. However, there are currently no specific classification algorithms tailored to this task. We trained a classification algorithm on 1,432 publicly available methylation profiles of a selection of 16 pediatric tumor entities and control tissues. To tune and evaluate its performance, we assembled a cohort of 67 pediatric CSF samples with a confirmed tumor diagnosis and a known presence of cfDNA. Our model scores a balanced accuracy of 67% on the test set, outperforming the currently used published models. This promising result highlights the potential of our approach in improving diagnostic accuracy for pediatric brain tumors.

Publikationsverlauf

Artikel online veröffentlicht:
09. Mai 2025