Parallel Adhesive Dynamics with Adaptive Physics Refinement for Large-Scale Tracking of Circulating Tumor Cells

Abstract

Cancer remains one of the leading causes of death worldwide, responsible for approximately 10 million deaths annually. Metastasis—the spread of cancer cells from a primary tumor to distant organs—is the primary driver of cancer-related mortality. Accurately predicting the trajectories of circulating tumor cells (CTCs) is crucial for advancing targeted anti-metastatic therapies. Increasing evidence indicates that CTC dynamics are largely influenced by local hemodynamic conditions, ligand-receptor interactions, and flow-induced wall shear stress (WSS). These biomechanical factors can be systematically analyzed using multi-physics models that couple computational fluid dynamics (CFD) with detailed cellular models. Such simulations provide critical insights into the mechanistic underpinnings of metastasis, providing a powerful framework for therapeutic innovation. Despite advances in modeling techniques, simulating complex metastasis processes remains a significant computational challengeExisting computational physics models of metastasis face several critical challenges. First, many adopt simplified assumptions—such as small, idealized microvascular geometries—that constrain their physiological relevance. Second, effective utilization of modern heterogeneous computing architectures requires simulation frameworks to be designed with performance portability across diverse hardware platforms. Third, elucidating the mechanistic links between flow-derived forces and metastatic progression requires efficient ensemble modeling techniques that can account for biological and biomechanical variability. Addressing these challenges is essential for advancing computational models as robust, predictive complements to experimental research.The objective of this thesis is to deepen our understanding of how fluid-derived forces—particularly wall shear stress (WSS)—govern the metastatic trajectories of circulating tumor cells (CTCs). To this end, a multi-scale, high-throughput, and performance-portable computational platform was developed to enable in silico tracking of CTCs at physiologically relevant tissue scales. Central to this effort was the systematic porting of our multiphysics modeling application, HARVEY, to multiple offload acceleration languages, ensuring compatibility across a wide range of high-performance computing architectures. Given the computational demands of high-resolution, large field-of-view simulations, performance portability was essential to effectively utilize heterogeneous hardware.To harness the capabilities of on-node graphics processing units (GPUs), the platform’s cellular communication algorithms were redesigned to minimize data movement, resulting in up to an order-of-magnitude reduction in simulation time-to-solution. To address memory bottlenecks associated with high-resolution grids used for modeling ligand-receptor interactions between CTCs and the endothelium, we developed a multi-scale adhesive dynamics algorithm featuring adaptive physics refinement (AD-APR). This approach enables long-distance tracking of adhesive CTCs while efficiently utilizing heterogeneous computing resources.The AD-APR framework was further extended to support ensemble simulations, enabling the simultaneous tracking of thousands of CTCs while significantly reducing storage requirements. To further optimize this approach, we developed a communication-free variant of the algorithm tailored for converged, steady-state flow conditions—scenarios commonly encountered in physiological and experimental settings.To explore the mechanistic influence of WSS on adhesive trajectory, the platform was augmented with a model of WSS-modulated adhesive dynamics. Using this enhanced framework, we demonstrate that the experimentally observed CTC attachment patterns can be recapitulated by incorporating receptor modulation as a function of local WSS.The work presented in this dissertation establishes a foundational framework for simulation-driven studies of metastasis, paving the way for translational research aimed at deepening biological insight and accelerating the discovery of next-generation anti-metastatic therapies.