A Novel Computational Architecture for Construction and Execution of Modular, Multiscale, Multi-Algorithmic Dynamical Models

We present a general-purpose, hierarchical modeling approach for integrating models of different portions of a dynamical system into a single, executable whole. The approach is particularly useful in cases where detailed models of small portions of a large, complex system have been constructed by multiple independent research teams using disparate modeling frameworks and specialized computer code. The technique relies on partitioning a system of interest into multiple functional modules that can operate relatively autonomously from each other for long periods of time relative to the timescales of their internal dynamics. Modules are organized in a hierarchical tree structure based on descending characteristic timescales and with a single top-level ("world") parent. Each module contains a "grey box" simulator engine commensurate with the modeling framework chosen for the local portion of the dynamical system. The interface between a simulator and the mulitscale architecture is limited to a collection of parameters and variables that are informed by, and passed to, neighboring modules. Data is transmitted between modules and simulations are executed and coordinated using a novel multiscale driving algorithm that is internal to the architecture. Direct intervention by the modeler is thus limited to (i) provision of appropriate simulators, (ii) presentation of relevant parameters and variables to the architecture, and (iii) definition of a small number of architecture-associated parameters. We present a formal mathematical justification for the proposed methodology and demonstrate the practical utility of the approach on a multiscale model of granulomar formation during Mycobacterium tuberculosis infection in the lung.