From Transcriptomics to Game-Theoretic Models of Treatment Resistance in Glioblastoma Organoids

Integrating patient-specific data with mechanistic mathematical models is essential for developing personalized cancer therapies. Here, we present a data-driven systems biology framework that combines transcriptomic analysis, machine learning, and game-theoretical modeling to predict evolutionary responses to irradiation in patient-derived glioblastoma organoids. Supervised and unsupervised learning approaches identify relevant cancer cell populations and quantify temporal changes in their composition from bulk transcriptomic data. These estimates parameterize replicator dynamics to infer transient evolutionary trajectories and evolutionary stable strategy (ESS) compositions. Our analyses reveal that hypoxia is the primary driver of short-term responses to irradiation, while exerting a smaller influence on long-term evolutionary outcomes. Furthermore, supervised approaches provide the most accurate estimates of evolutionary dynamics when therapy resistance is represented as a qualitative trait. This integrative framework demonstrates how transcriptomic data can inform mechanistic models of tumor evolution, supporting patient-specific treatment optimization and advancing data-driven mathematical oncology.