Agent-based modeling is a discrete-event, object-oriented, spatially-explicit type of computer simulation that is an increasingly popular modeling method for converting the correlations identified from Big Data into dynamic representations of mechanistic knowledge. Agent-based models (ABMs) representing populations of interacting cells have been used to examine a range of physiological/pathophysiological systems such as cancer, sepsis, infectious disease, wound healing, and gastrointestinal disease. A natural step in the evolution of agent-based modeling is the desire to develop high-resolution, anatomic-scale organ ABMs that can reproduce recognizable clinical pathophysiology. However, the operational challenge of effectively parameterizing (calibrating), characterizing (meta-modeling) and validating such models are daunting, if not computationally intractable as a practical issue, given existing methods. To help address this issue, we propose to utilize automated adaptive simulation workflows on anatomic-level multi-scale agent-based models to enable and make tractable the process of exploring the parameter and behavior spaces of very large (hundreds of billions of agents) ABMs able to represent entire organ systems. These workflows, already tested in the characterization of smaller scale ABMs, will be extended to state-of- the-art high performance computing (HPC) environments in order to demonstrate and eventually provide this capability to researchers developing larger and more complex ABMs, fundamentally changing how such models are used and analyzed. We will utilize a high-performance version of the Swift task-parallel scripting language (Swift/T), to perform parameterization and behavior-space exploration of an enhanced version of the Spatially Explicit General-purpose Model of Enteric Tissue (SEGMEnT) HPC implemented at anatomic scale, i.e., the entire small and large intestine. This project includes performing parameter-space characterization of SEGMEnT HPC at multiple scaling levels using previously identified objective functions derived from tissue- and organ-level features of intestinal tissue, porting SEGMEnT HPC to Repast HPC, an existing HPC-capable ABM toolkit. In doing so we will expand Repast HPC's ability to represent complex biological phenomena, and develop adaptive simulation workflows using Swift/T and Repast HPC, tested in the Repast HPC implementation of SEGMEnT, in order to facilitate parameter space exploration and characterization by the general modeling community.
Agent-based modeling is increasingly being used to perform modeling and simulation of multi-scale biological processes. However, one of the limitations and barriers to the wider and more effective use of agent-based models (ABMs) is the paucity of formal methods for their analysis, with significant impact on the parameterization, calibration, validation and utility of increasingly large and complex anatomic-scale ABMs. We propose that the implementation of automated adaptive simulation workflows performed on high-performance computing environments can be used to more effectively, efficiently and comprehensively characterize very large (i.e. billions of computational agents) ABMs. This project will utilize the Repast HPC agent-modeling toolkit and the Swift/T parallel-task management programming language and runtime environment to develop automated adaptive simulation workflows for a very large ABM of the intestinal tract, the Spatially Explicit General-purpose Model of Enteric Tissue (SEGMEnT) HPC. The adaptive simulation workflows developed will serve as a reference application of this method, and serve as an initial step towards developing a scalable and extensible strategy for analyzing very large ABMs.
