Establishing Digital Twins for High Throughput Cellular Analysis in Whole Blood

Understanding how cancer spreads through the body requires tracking how circulating tumor cells (CTCs) move through the bloodstream and interact with vessel walls. These adhesive interactions occur at submicrometer scales, making them incredibly difficult and expensive to simulate in anatomically realistic settings. To address this challenge, researchers at Duke University are using ALCF computing resources to develop an adaptive, GPU-accelerated modeling framework that captures high-resolution cancer cell adhesion behavior across large vascular domains.

Challenge

Simulating the adhesive dynamics (AD) of cancer cells requires resolving fine-scale molecular interactions between CTC ligands and receptors on blood vessel walls, while also modeling fluid flow over millimeter- to centimeter-scale vascular geometries. Performing such high-fidelity simulations across large domains is infeasible with traditional methods due to the extreme memory and computational demands. Most existing models are limited to small, idealized vessels and cannot scale to anatomically realistic systems. A new approach is needed to bridge the gap between microscale biological accuracy and macroscale anatomical context.

Approach

The team extended a technique known as adaptive physics refinement (APR), which dynamically couples a finely resolved simulation window to a larger coarse-grained fluid domain. In this work, they advanced the APR framework to allow the window to traverse vessel walls, enabling detailed simulation of ligand-receptor binding events at the cell surface. The model integrates lattice Boltzmann fluid dynamics with stochastic adhesive dynamics and is optimized for GPU execution. Researchers used the ALCF’s Aurora supercomputer to run large-scale simulations with significantly reduced memory costs, leveraging heterogeneous CPU-GPU resources and multi-resolution spatial modeling.

Results

The APR-AD framework accurately reproduced adhesive interactions between a cancer cell and patterned wall receptors in a complex microfluidic geometry. Simulations showed strong agreement with fully explicit, high-resolution models while requiring far fewer resources—reducing memory usage by more than 25 times. Performance optimizations, including GPU-accelerated random number generation and octree-based receptor searches, significantly sped up the most computationally intensive portions of the workflow. Validation tests confirmed that the framework preserves both flow accuracy and bond formation behavior across a range of simulation conditions.

Impact

This work provides a scalable and efficient method for simulating cancer cell adhesion at anatomical scales, enabling more detailed exploration of how CTCs interact with the vascular system during metastasis. By reducing computational costs without sacrificing biological accuracy, the APR-AD framework opens new possibilities for simulating cancer progression and testing therapeutic strategies. The approach also demonstrates how advanced multiscale modeling techniques and leadership-class computing can be combined to tackle complex problems in biomedical engineering.