Developing a pediatric pathology foundation model for morphological and epigenetic biomarker discovery in childhood cancers

The National Cancer Institute awarded the University of Pennsylvania $100,228 on January 27, 2027, under the Cancer Research Manpower program (CFDA 93.398) to develop a pediatric pathology foundation model for morphological and epigenetic biomarker discovery in childhood cancers.

The project will build the first pan-cancer pediatric pathology foundation model using more than 800,000 whole-slide images from approximately 70,000 pediatric patients at the Children's Hospital of Philadelphia, then integrate whole-slide image and DNA methylation data to quantify the prognostic value of histomics combined with methylomics in pediatric gliomas. The deliverables include computational frameworks that identify clinical contexts in which histology alone suffices for accurate risk stratification versus those requiring epigenetic profiling, with the intent to provide clinicians and pathologists with tools supporting personalized clinical management of childhood cancers. The work addresses the gap between pediatric and adult cancer biomarker development, where current diagnostic and prognostic evaluation relies on subjective interpretation of hematoxylin and eosin-stained images and DNA methylation profiling—approaches not easily accessible in certain clinical settings.

Work is performed in Philadelphia, Pennsylvania, with an ultimate completion date of July 31, 2029.

Abstract

Pediatric cancers are the leading cause of disease-related death among children and adolescents in the united states. Despite having unique genetic, epigenetic, and histologic profiles compared to adult cancers, childhood malignancies remain comparatively understudied. As a result, current frameworks for clinical assessment often rely on biomarkers identified in adult cohorts, limiting their relevance and utility in pediatrics. These limitations are particularly evident in pediatric gliomas, where diagnostic and prognostic evaluation relies on subjective interpretation of hematoxylin and eosin (h & e)-stained whole-slide images (wsis) and DNA methylation profiling, which are not easily accessible in certain clinical settings. This proposal aims to develop scalable and interpretable computational frameworks that leverage recent advances in digital pathology and multimodal integration to improve tumor characterization and survival prediction in pediatric oncology. In aim 1, we will build the first pan-cancer pediatric pathology foundation model using more than 800,000 wsis from approximately 70,000 pediatric patients at the children's hospital of philadelphia. In aim 2, we will integrate wsis and DNA methylation data to quantify the prognostic value of histomics when integrated with methylomics in pediatric gliomas to identify clinical contexts in which histology alone is sufficient and those in which epigenetic profiling is required for accurate risk stratification. This proposal will establish reproducible, interpretable, and accessible computational frameworks tailored to pediatric oncology, providing clinicians and pathologists with tools to support personalized clinical management of childhood cancers.