St. Jude AI Liquid Biopsy Classifies Pediatric Brain Tumors with 92% Accuracy

🧠 The Growing Challenge of Pediatric Brain Tumors

Pediatric brain tumors represent the most common type of solid tumor cancer diagnosed in children, affecting approximately 5,000 new cases annually in the United States alone. These abnormal growths in the brain or central nervous system can disrupt vital functions, leading to symptoms like headaches, seizures, nausea, and developmental delays. Unlike adult brain tumors, which often metastasize from other cancers, pediatric versions typically originate within the brain tissue itself.

Key types include medulloblastomas, which arise in the cerebellum and account for about 20% of cases; gliomas, originating from glial cells that support neurons; ependymomas from ependymal cells lining the ventricles; and rarer forms like diffuse intrinsic pontine gliomas (DIPG), notorious for their location in the brainstem, making surgery nearly impossible. Survival rates vary widely: medulloblastomas have five-year survival around 70-95% depending on molecular subgroups, while high-grade gliomas hover at 20-30%. Overall, brain tumors are the leading cause of cancer-related death in children aged 0-14, underscoring the urgent need for better diagnostics and treatments.

Diagnosis traditionally relies on magnetic resonance imaging (MRI) scans to locate tumors, followed by surgical biopsy for molecular profiling. However, biopsies carry significant risks in young patients, including bleeding, infection, neurological deficits, and challenges in accessing deep-seated tumors. Accurate molecular classification, as per World Health Organization (WHO) guidelines, is crucial since it determines prognosis and therapy—yet obtaining tissue samples remains invasive and sometimes infeasible.

Overcoming Diagnostic Hurdles with Liquid Biopsies

Liquid biopsies offer a promising non-invasive alternative by analyzing circulating tumor DNA (ctDNA)—genetic fragments shed by tumors into bodily fluids like blood or cerebrospinal fluid (CSF). In pediatric brain tumors, ctDNA levels in blood are often too low due to the blood-brain barrier, making CSF the preferred medium. Obtained via lumbar puncture, CSF circulates around the brain and spinal cord, capturing tumor-derived DNA more effectively.

Advantages include repeatability for monitoring treatment response or detecting relapse without repeated surgeries, real-time insights into tumor evolution, and minimal risk compared to craniotomy. Challenges persist: ctDNA fractions in CSF can be as low as 0.1-1%, drowned out by normal cell-free DNA (cfDNA) from healthy cells. Traditional classifiers, designed for high-DNA tissue samples, falter here, necessitating innovative approaches like artificial intelligence (AI) to discern subtle methylation patterns—epigenetic modifications acting as tumor 'fingerprints.'

• Enables serial sampling during therapy
• Reduces procedural risks for fragile pediatric patients
• Supports precision medicine by identifying actionable mutations

Recent advances have validated CSF liquid biopsies for detecting mutations in gliomas and medulloblastomas, paving the way for tools that classify tumors molecularly without tissue.

St. Jude's M-PACT: A Game-Changing AI Tool

Published on February 17, 2026, in Nature Cancer, researchers from St. Jude Children's Research Hospital, led by Paul Northcott, PhD, introduced M-PACT (Methylation-based Predictive Algorithm for CNS Tumors). This AI-powered platform classifies pediatric brain tumors using minute ctDNA amounts from CSF, achieving 92% accuracy against tissue benchmarks. Co-first authors Katie Han and Kyle Smith, PhD, collaborated with international teams from Heidelberg, Vienna, Amsterdam, and Tampere.

"This is a next-generation assay and computational framework optimized for pediatric neuro-oncology," Northcott stated, highlighting its versatility across clinical scenarios. Developed in St. Jude's Center of Excellence in Neuro-Oncology Sciences, M-PACT builds on decades of progress, where childhood cancer survival rose from 20% to 80% since the hospital's founding.

🔬 How M-PACT Works: From Data to Diagnosis

M-PACT employs a deep neural network trained on over 5,000 DNA methylation profiles from approximately 100 tumor entities. Unlike conventional methods tuned for tissue, it simulates real-world CSF conditions by computationally blending tumor methylation data with normal cfDNA datasets, teaching the AI to detect ctDNA signals amid noise.

The process unfolds in steps:

1. Sample Collection: A small CSF volume via lumbar puncture yields cfDNA.
2. Methylation Profiling: Bisulfite sequencing reveals epigenetic patterns unique to tumor types.
3. AI Classification: The model analyzes patterns, outputting tumor entity, grade, and molecular subtype.
4. Microenvironment Insights: Predicts non-tumor cell fractions (e.g., T cells, B cells) influencing therapy response.
5. Monitoring: Serial tests track changes in tumor aggression or relapse risk.

"We reversed the usual flow, designing M-PACT for ctDNA first," Han explained. This innovation addresses low-input challenges, matching or exceeding tissue biopsy standards.

For context, DNA methylation involves adding methyl groups to cytosine bases, regulating gene expression without altering sequence. Tumors exhibit distinct profiles, enabling precise subtyping like SHH-activated medulloblastoma versus Group 3.

Study Results: Proven Accuracy and Utility

Validated across multinational cohorts, M-PACT identified 92% of brain tumors correctly, even with tiny ctDNA. It distinguished true relapses from secondary malignancies—critical since treatments differ—and monitored therapy effects, revealing shifts in tumor microenvironment under pressure.

"Even tiny ctDNA amounts can be accurately classified," Smith noted. In proof-of-concept cases, it diagnosed tumors at surgery, tracked post-treatment evolution, and predicted immune cell infiltration, informing immunotherapy decisions.

Beyond classification, it forecasts how cancers recruit normal cells, offering new therapeutic angles. For more on St. Jude's innovations, explore their press release.

Clinical Implications for Patients and Providers

M-PACT sets a new diagnostic standard, enabling faster, safer molecular profiling essential for risk-stratified therapy. Low-risk tumors might avoid aggressive chemo-radiation; high-risk ones gain targeted drugs like BET inhibitors for certain medulloblastomas.

For families, it means less trauma: lumbar punctures replace open surgery. Providers benefit from real-time data, adjusting protocols dynamically. In resource-limited settings, its low-input efficiency democratizes advanced diagnostics.

Integration with platforms like the Pediatric Brain Tumor Portal accelerates research. Learn about ongoing trials via St. Jude's research jobs listings.

Future Directions and Broader Impact

Northcott envisions expanding M-PACT to all pediatric cancers, including solid tumors and leukemias. "The informatics will grow," he said, with community adoption likely. Challenges like standardization and prospective trials remain, but its foundation is robust.

This 'team science' exemplifies global collaboration, blending bioinformatics, pathology, and oncology. For academics pursuing such work, opportunities abound in higher education faculty positions or clinical research jobs.

Photo by Abdulai Sayni on Unsplash

Advancing Pediatric Oncology Through Research Careers

Innovations like M-PACT highlight the demand for experts in neuro-oncology, AI, and genomics. Institutions seek professors and postdocs to drive precision medicine. Explore postdoc opportunities or professor jobs to contribute.

Share your insights in the comments below—your professor's expertise might inspire the next breakthrough. Check Rate My Professor for guidance, or browse higher ed jobs and university jobs for roles shaping tomorrow's cures. AcademicJobs.com connects you to these vital positions.