AI-Powered Liquid Biopsy Classifies Pediatric Brain Tumors with High Accuracy

Liquid biopsies offer a noninvasive way to study cancer by analyzing circulating tumor DNA in body fluids. However, in pediatric brain tumors, the small amount of ctDNA in cerebrospinal fluid has limited the effectiveness of current diagnostic tools. Traditional methylation-based classifiers designed for tissue samples often fail when applied to liquid biopsies. Researchers have now developed an artificial intelligence (AI)-powered system that accurately classifies pediatric brain tumors using ctDNA, improving diagnosis, monitoring, and relapse detection.

Scientists at St. Jude Children’s Research Hospital (Memphis, TN, USA, in collaboration with international centers, have developed the Methylation-based Predictive Algorithm for CNS Tumors (M-PACT), an AI framework designed specifically for ctDNA analysis. M-PACT employs a deep neural network trained on more than 5,000 DNA methylation profiles across approximately 100 tumor types. By computationally integrating tumor reference datasets with normal cell-free DNA profiles, the system was optimized to detect and classify even very small amounts of ctDNA.

In benchmarking tests, M-PACT correctly identified 92% of pediatric brain tumors using cerebrospinal fluid samples. The system could also distinguish true relapse from secondary malignancies and track tumor progression or treatment response without additional input. The findings, published in Nature Cancer, demonstrate that methylation-based ctDNA analysis can match or exceed the diagnostic standards of tissue biopsy in certain contexts.

In addition to tumor classification, M-PACT can analyze the tumor microenvironment by identifying contributions from immune and other noncancerous cells within cerebrospinal fluid. This capability enables monitoring of disease evolution during therapy, when tissue sampling is rarely performed. Although initially developed for pediatric brain tumors, researchers believe the platform can be adapted to other solid tumors and hematological malignancies. Ongoing efforts aim to expand the informatics framework to encompass a broader range of childhood cancers.

“M-PACT provides us with a new lens to monitor disease evolution, especially during therapy, when tissue sampling isn’t typically done,” said co-first author Katie Han. “Now we can start to see how both the tumor and its microenvironment change with therapeutic pressure.”

Related Links:
St. Jude Children’s Research Hospital