In a pioneering advancement that promises to reshape pediatric neuro-oncology, researchers at St. Jude Children’s Research Hospital, in concert with international collaborators, have unveiled an artificial intelligence-driven platform named M-PACT (Methylation-based Predictive Algorithm for CNS Tumors). This innovative tool leverages liquid biopsies, specifically analyzing circulating tumor DNA (ctDNA) gathered from cerebrospinal fluid, to molecularly classify pediatric brain tumors with unparalleled precision. Published in the esteemed journal Nature Cancer, M-PACT introduces a transformative approach to diagnosing, monitoring, and surveilling brain tumors in children, overcoming long-standing technical challenges caused by the scant amounts of ctDNA available in these cases.
Liquid biopsies have long been celebrated for their noninvasiveness and ability to access genomic information without the need for risky surgical interventions. However, their utility in pediatric brain tumors has been constrained by the limited quantity of ctDNA present in cerebrospinal fluid, rendering traditional diagnostic techniques insufficient. M-PACT addresses this bottleneck by employing a deep neural network trained on an extensive collection of over 5,000 DNA methylation profiles representing roughly 100 tumor entities. This massive training dataset enabled the algorithm to discern subtle, disease-specific methylation patterns that are exquisitely informative yet often masked in low-input ctDNA samples.
One of the central innovations behind M-PACT is its deliberate shift in design philosophy. Unlike prior classifiers that were primarily optimized for tumor tissue specimens, where DNA input is more abundant, this AI framework is tailor-made for the low-input ctDNA from liquid biopsies. Co-first author Katie Han, a PhD student at St. Jude, emphasizes that this inversion is critical — M-PACT was designed around ctDNA’s unique characteristics and subsequently demonstrated applicability to tissue, marking a paradigm reversal in diagnostic development. Through computationally merging large tumor methylation reference data with datasets of normal cell-free DNA, the team achieved an algorithm that reliably classifies tumors from minimal biological material.
Benchmarking tests of M-PACT exhibited promising performance, accurately identifying 92% of pediatric brain tumors from cerebrospinal fluid samples alone. Beyond initial diagnosis, the platform exhibits dynamic capabilities such as discerning tumor relapse from secondary malignancies, and importantly, monitoring tumor progression or regression in response to therapy. This capacity allows clinicians unprecedented real-time insights into disease evolution during treatment, without demanding additional invasive sampling—a critical advantage given the delicate nature of pediatric patients and the potential risks of repeat biopsies.
M-PACT’s power extends beyond tumor cell identification. Its sensitivity enables it to detect noncancerous cellular components within the cerebrospinal fluid, revealing intricate details about the tumor microenvironment—a component increasingly recognized as vital in cancer progression and therapeutic resistance. By quantifying DNA contributions from immune cells like T and B lymphocytes, the technology affords a novel perspective on how cancers manipulate their surrounding cellular milieu. This nuanced insight offers significant potential for investigating tumor-host interactions and could open new therapeutic avenues targeting the microenvironment.
From a translational perspective, the utility of M-PACT is formidable. At the time of surgical intervention, the algorithm can make precise tumor classifications using only cerebrospinal fluid, facilitating timely and accurate treatment decisions. Moreover, during follow-up care, M-PACT can autonomously indicate whether a recurring tumor represents a true relapse or a newly arisen secondary tumor, information crucial for directing appropriate clinical strategies. These capabilities collectively establish M-PACT as a next-generation tool with the potential to revolutionize the pediatric neuro-oncology diagnostic workflow.
The interdisciplinary innovation behind M-PACT was driven by robust international collaboration. Partners from the Hopp Children’s Cancer Center Heidelberg, German Cancer Research Center, Medical University of Vienna, Amsterdam University Medical Centers, and multiple other institutions contributed critical data and expertise, enabling the assembly of one of the most comprehensive clinically annotated pediatric liquid biopsy cohorts to date. This cooperative framework underscored the study’s success, blending computational science, molecular biology, and clinical expertise across continents to solve complex biomedical challenges.
From a technical standpoint, M-PACT’s use of deep learning relies on supervised neural network architectures that ingest methylation signatures characteristic of distinct tumor types. This advanced computational strategy enables the model to handle heterogeneous input sources and minute DNA quantities, extracting discriminatory features masked in traditional analyses. As a result, M-PACT surpasses prior methylation classifiers that were predominantly crafted for solid tumor tissue, setting a high bar for sensitivity and specificity in minimally invasive diagnostics.
Dr. Paul Northcott, who spearheaded the research at St. Jude’s Center of Excellence in Neuro-Oncology Sciences, highlighted the broad future implications of this technology. While initially validated in pediatric brain tumors, he predicts that the underlying informatics framework will expand to encompass a wider array of pediatric and adult malignancies, including hematologic cancers and solid tumors elsewhere in the body. This adaptability marks M-PACT as a foundational tool poised for broad diagnostic and therapeutic impact.
Funding for this groundbreaking project was secured from a diverse coalition of agencies, including prominent cancer research foundations, international health organizations, and government bodies such as the National Cancer Institute and the Finnish Ministry of Social Affairs and Health. This broad support reflects the significance of advancements like M-PACT in enhancing cancer diagnostics and improving patient outcomes globally.
In sum, M-PACT embodies a quantum leap in leveraging artificial intelligence to amplify the diagnostic power of liquid biopsies in pediatric brain tumors. Its sophisticated molecular classification capacity, coupled with its ability to delve into tumor microenvironment dynamics, heralds a new era where clinicians gain comprehensive, real-time insights into tumor identity and behavior through minimally invasive means. As research and clinical applications progress, M-PACT stands to transform pediatric oncology care, reducing reliance on invasive procedures and ultimately improving survival and quality of life for young cancer patients worldwide.
Subject of Research: Pediatric brain tumor classification using liquid biopsies and AI analysis of circulating tumor DNA methylation patterns.
Article Title: Classifying pediatric brain tumors by liquid biopsy using artificial intelligence
News Publication Date: February 17, 2026
Web References:
10.1038/s43018-026-01115-4
Image Credits: St. Jude Children’s Research Hospital
Keywords: Brain tumors, Liquid biopsies, Circulating tumor DNA, Cerebrospinal fluid, DNA methylation, Pediatric neuro-oncology, Artificial intelligence, Deep learning, Tumor microenvironment, Cancer diagnostics, ctDNA classification, Tumor relapse detection
