An AI-powered liquid biopsy tool called M-PACT has achieved 92% accuracy in classifying pediatric brain tumors using circulating tumor DNA (ctDNA) from cerebrospinal fluid. Developed by scientists at St. Jude Children's Research Hospital in collaboration with researchers from the Hopp Children's Cancer Center Heidelberg (KiTZ), the German Cancer Research Center (DKFZ), and other international institutions, M-PACT represents a major leap forward in noninvasive cancer diagnostics. Liquid biopsies offer a less invasive alternative to traditional tissue biopsies by analyzing ctDNA shed by tumors into body fluids. However, pediatric brain tumors typically release very small amounts of ctDNA, making detection and classification extremely challenging. To overcome this, the research team created M-PACT, a methylation-based predictive algorithm that uses artificial intelligence to analyze DNA methylation patterns—chemical markers that influence gene activity and serve as unique identifiers for different tumor types. Unlike previous methods that relied on classifiers designed for tissue samples with abundant DNA, M-PACT was specifically built for the low-input conditions of liquid biopsies. The system was trained on over 5,000 DNA methylation profiles across nearly 100 tumor types, using a deep neural network that combines large reference datasets with normal cell-free DNA data. This innovative training approach enables accurate tumor classification even when ctDNA is scarce. In benchmark testing, M-PACT correctly identified 92% of pediatric brain tumors. It can also distinguish between relapse and secondary cancers, track tumor progression or response to treatment, and monitor changes in the tumor microenvironment—such as shifts in immune cell populations like T cells and B cells—without requiring additional data. “This is a next-generation assay and computational framework,” said Paul Northcott, Ph.D., corresponding author and director of the Center of Excellence in Neuro-Oncology Sciences at St. Jude. “M-PACT takes liquid biopsy to another level in pediatric neuro-oncology, enabling diagnosis, monitoring, and surveillance in ways that were previously impossible.” The tool’s ability to analyze the “negative space” of the tumor—noncancerous cells in the fluid—opens new avenues for understanding how tumors manipulate their surroundings. This insight could reveal how therapies affect both cancer cells and their supporting microenvironment, offering a more complete picture of disease evolution. While initially developed for pediatric brain tumors, the researchers believe M-PACT’s framework has broad potential across many cancer types, including other solid tumors and blood cancers. The team is already working to expand the system’s capabilities to cover the full spectrum of childhood cancers. The success of M-PACT underscores the power of international collaboration. The project brought together experts in computational biology, oncology, and data science from multiple institutions, demonstrating how shared expertise and data can drive transformative innovation. “This study is a prime example of team science,” Northcott said. “When we combine complementary skills and work toward a common goal, we can achieve what no single lab could do alone.”