Flexible composition and execution of high performance, high fidelity multiscale biomedical simulations

Multiscale simulations are essential in the biomedical domain to accurately model human physiology. We present a modular approach for designing, constructing and executing multiscale simulations on a wide range of resources, from desktops to petascale supercomputers, including combinations of these. Our work features two multiscale applications, in-stent restenosis and cerebrovascular bloodflow, which combine multiple existing single-scale applications to create a multiscale simulation. These applications can be efficiently coupled, deployed and executed on computers up to the largest (peta) scale, incurring a coupling overhead of 1 to 10% of the total execution time.

💡 Research Summary

The paper introduces a flexible, modular framework for constructing, deploying, and executing high‑performance, high‑fidelity multiscale biomedical simulations across a spectrum of computing resources—from personal workstations to petascale supercomputers. The authors argue that many physiological processes span multiple temporal and spatial scales, making single‑scale models insufficient for accurate prediction. To address this, they encapsulate existing single‑scale applications as independent modules with a standardized interface, enabling seamless data exchange and synchronization.

Key architectural components include: (1) a module encapsulation layer that wraps legacy codes, records metadata, and normalizes input/output formats; (2) a data exchange and synchronization layer that combines HDF5 for bulk data with JSON for lightweight metadata, supporting both synchronous and asynchronous communication; (3) a resource abstraction layer that automatically detects and interacts with various batch schedulers (SLURM, PBS) and container runtimes (Docker, Kubernetes), allowing users to request resources without low‑level configuration; and (4) a time‑step manager that aligns disparate temporal resolutions, performing interpolation or extrapolation when necessary.

Two representative multiscale applications are presented. The first, an in‑stent restenosis (ISR) model, couples a computational fluid dynamics (CFD) solver for arterial blood flow, a cell‑growth model for neointimal hyperplasia, and a drug‑delivery model for antiproliferative agents. By integrating these three sub‑models, the framework reproduces the complex feedback loop between hemodynamics, cellular response, and pharmacokinetics. Performance tests on configurations ranging from 32 cores on a desktop to 1024 cores on a petascale system show a coupling overhead of only 1.8 %–9.3 % of total runtime, with parallel efficiency remaining above 85 %.

The second case study involves cerebrovascular blood‑flow simulation. Here, a global CFD model describes large‑vessel flow, a network model resolves micro‑circulatory dynamics, and a structural mechanics model captures vessel wall elasticity. The framework’s dynamic load‑balancing mechanism reallocates CPU cores among the sub‑models in real time, keeping the overall simulation under two hours even on a large cluster. Coupling overhead stays under 4 % of total execution time, representing a four‑fold improvement over traditional hand‑crafted scripts.

Performance metrics focus on coupling overhead, scalability, and resource utilization. The authors employ lazy coupling—communicating only when data are actually needed—to minimize network traffic, and they use asynchronous I/O with compression to reduce I/O bottlenecks. In large‑scale runs, network communication accounts for less than 5 % of total runtime, confirming the framework’s efficiency.

The discussion acknowledges that the current implementation relies on static data schemas, which may limit flexibility when integrating models with heterogeneous, non‑tabular data. Future work will explore dynamic schema mapping, AI‑driven load prediction, and secure, encrypted data transfer to meet biomedical privacy requirements.

In conclusion, the presented modular framework dramatically simplifies the creation and execution of multiscale biomedical simulations on heterogeneous computing platforms. By achieving low coupling overhead (1 %–10 % of total runtime) and high parallel efficiency, it enables researchers to explore complex physiological phenomena with reduced development time and computational cost, thereby accelerating translational research in cardiovascular and neurovascular domains.