Beyond static biomarkers: systems biology and AI for decoding cancer dynamics

Precision oncology has been built largely on static biomarkers, including mutational profiles, receptor status, histopathologic classes, and single-time-point molecular signatures. These readouts have transformed diagnosis and treatment selection, but they remain mismatched to a disease whose most consequential behaviors-progression, metastasis, treatment adaptation, dormancy, and relapse-are dynamic, noisy, and multiscale. In this article, we argue that cancer is better understood as a stochastic, coupled biological system than as a fixed molecular identity. We bring together adjacent literatures spanning cancer cell states, spatial and ecological organization, metabolism and dormancy, mechanobiology, longitudinal biomarkers, quantitative oncology, and AI-enabled representation learning, and develop a mathematically grounded framework for reasoning about cancer dynamics. Our central claim is modest but consequential: future biomarkers should not only classify current disease state, but also estimate transition risk, system instability, and trajectory direction. To support that claim, we distinguish latent biological state from clinical observation, clarify why partial observability makes dynamic inference difficult, and introduce a mathematically explicit but deliberately constrained formal scaffold based on stochastic state-space models, local linearization, and layer-specific dynamical motifs. We then develop six internal biological layers of cancer dynamics-molecular regulatory dynamics, cellular state plasticity, spatial niche organization, tumor ecosystem co-evolution, metabolic-epigenetic coupling with dormancy, and mechanobiological feedback-and treat longitudinal clinical monitoring as a linked observation layer rather than a mechanistic subsystem. Particular attention is given to noise; transcriptional noise, ecological variability, treatment-induced perturbation, and measurement noise all shape how cancer states are occupied, destabilized, and detected. Finally, we review dynamic biomarker evidence from ctDNA-guided adjuvant therapy, circulating tumor cells, serial imaging, and adaptive therapy, discuss how AI can support inference under partial observability without replacing mechanism, examine regulatory and health-equity constraints, and outline the research agenda needed to turn dynamic oncology from a compelling idea into a reproducible clinical discipline.