Related Experiment Video
Updated: Oct 1, 2026
Digital Spatial Profiling for Characterization of the Microenvironment in Adult-Type Diffusely Infiltrating Glioma
Published on: September 13, 2022
Machine learning and computational approaches to model therapeutic response and resistance in diffuse midline glioma
Océane A Dubois1,2,3, Olive E L Loughnan1,2,3, Yuanhao Jiang1,2,3
1Cancer Signalling Research Group, School of Biomedical Sciences & Pharmacy, College of Health, Medicine and Wellbeing, University of Newcastle, Callaghan, NSW, Australia.
Purpose:
Diffuse midline glioma (DMG) remains one of the most lethal cancers affecting children, adolescents and young adults and is near-universally resistant to treatment. Histone H3-alterations establish a profoundly dysregulated epigenetic landscape that promotes extensive intratumoral heterogeneity and cellular plasticity, key drivers of therapeutic resistance and treatment failure. Although single-cell RNA sequencing and spatial transcriptomics have transformed the study of tumor evolution, the mechanisms underpinning treatment resistance in DMG remain poorly understood. This Review examines how computational and machine learning-based approaches can be leveraged to study tumor adaptation under therapeutic pressure.
Methods:
We provide an overview of computational frameworks developed for the analysis of single-cell and spatial transcriptomic datasets to model four major axes of tumor evolution: i) compositional shifts, ii) functional state remodeling, iii) tumor plasticity, and iv) intercellular communication; while highlighting how these approaches provided novel insights into DMG biology and the mechanisms underlying treatment adaptation.
Results:
Computational and machine learning-based frameworks provide powerful tools for modeling the spatiotemporal dynamics of tumor evolution in high-dimensional transcriptomic data. Across the four dimensions examined, these approaches identify resistant cellular populations, characterize adaptive transcriptional programs, reconstitute cell-state transitions, and map tumor-microenvironment interactions, revealing mechanisms of therapeutic resistance and treatment adaptation.
Conclusion:
Computational approaches developed to model tumor evolution under therapeutic pressure using single-cell and spatial transcriptomic data are rapidly advancing. As increasingly large and multimodal DMG datasets become available, the application of these approaches may help uncover resistance mechanisms, therapeutic vulnerabilities, and biomarkers that inform more precise and patient-specific treatment strategies.
