EMEWS

EMEWS enables large-scale model exploration and high-throughput computational experiments on high-performance computing systems to support parameter-space exploration, calibration, and optimization of agent-based models, microsimulations, and machine learning model training.

Key Features:

• Scalability: Executes millions of model runs on major high-performance computing (HPC) systems for high-throughput experiments.
• Multi-language integration: Integrates model exploration algorithms implemented in multiple programming languages to support heterogeneous algorithm use.
• Swift/T parallel scripting: Built on the Swift/T parallel scripting language for parallel workflow composition and task scheduling.
• Black-box model support: Supports "black box" applications including agent-based models, microsimulations, and machine learning model training that require numerous iterations.
• PhysiCell integration: Integrates with PhysiCell for 3-D multicellular simulation and high-dimensional parameter exploration.
• Adaptive sampling and optimization: Supports active learning and genetic algorithms for adaptive sampling and optimization of model parameters.
• Calibration with IMABC: Implements Incremental Mixture Approximate Bayesian Computation (IMABC) for calibration of microsimulation models such as CRC‑SPIN 2.0.

Scientific Applications:

• Cancer research: Explores cancer-host interactions and multiscale dynamical systems to systematically investigate therapeutic outcomes, including integration with PhysiCell for 3-D simulations.
• Immunotherapy design spaces: Performs dynamic exploration of immunotherapy design spaces using detailed dynamical simulation models combined with active learning and genetic algorithms to identify parameter regimes for cancer regression.
• Microsimulation model calibration: Calibrates microsimulation models (MSMs), including CRC‑SPIN 2.0, using IMABC to improve predictive accuracy for population-level outcomes and policy guidance.
• Machine learning model training: Enables large-scale training and evaluation of machine learning models as part of model exploration and optimization workflows.

Methodology:

Uses high-throughput computing frameworks and the Swift/T parallel scripting language to manage and execute extensive computational experiments; integrates external simulators such as PhysiCell and employs active learning, genetic algorithms, and Incremental Mixture Approximate Bayesian Computation (IMABC) for parameter-space exploration, adaptive sampling, optimization, and calibration based on data-driven error metrics.