Validation of a liquid biopsy based molecular diagnostic toolkit for pediatric sarcomas

This project tackled a major challenge in diagnosing childhood cancers using liquid biopsies - simple blood tests that analyse fragments of tumor-derived DNA released into the bloodstream. Many paediatric cancers have very few genetic mutations, which makes them difficult to detect with current liquid biopsy assays that rely on genetic aberrations only. To overcome this, we developed DEFT (Deep-learning-based Fragment-level Tumor detector), an artificial intelligence model that combines multiple types of information from cell-free DNA (cfDNA): how the DNA is cut into fragments, its chemical (methylation) patterns, and its position in the genome. By analyzing each DNA fragment individually, DEFT can determine whether it came from a tumor or from healthy tissue. DEFT proved to be more accurate than any other method tested for detecting tumor DNA in the blood of children with Ewing sarcoma. Importantly, it also allowed us to identify which patients are at low risk and respond very well to treatment. This could make it possible to safely reduce the intensity of chemotherapy for these children, protecting them from long-term side effects that currently reduce their healthy life expectancy. More broadly, DEFT offers the first AI-based solution to a fundamental question in liquid biopsy research: Which DNA fragments in a blood sample truly come from the tumor? Knowing this helps doctors detect cancer more reliably and gives researchers new insights into how tumor DNA enters the bloodstream. To bring this technology closer to real-world medical use, we also developed a new testing method based on nanopore sequencing, a fast, low-cost, and easy-to-use sequencing platform. This will make it more practical for hospitals and diagnostic laboratories to adopt DEFT and perform advanced liquid biopsy analyses. Overall, this project provides a powerful new tool for improving cancer diagnosis and monitoring in children - using a simple blood sample and without relying on genetic mutations.