Reproducible parallel inference and simulation of stochastic state space models using odin, dust, and mcstate.

Area of Science:

• Computational Biology
• Epidemiology
• Statistical Modeling

Background:

• State space models, including compartmental models, are vital for simulating physical, biological, and social systems.
• Sequential Monte Carlo methods are commonly used for parameter inference from time-series data.
• Engineering challenges hinder routine application of these methods, including adaptable model design, code performance, and statistical integration.

Purpose of the Study:

• To present a suite of R packages designed to simplify the creation and deployment of state space models.
• To address engineering challenges in model design, performance, and statistical technique integration.
• To provide tools beneficial for infectious disease modelers and adaptable to other scientific domains.

Main Methods:

• Development of R packages that convert user-defined models into parallelized C++ code.
• Integration of a fast, parallel, and reproducible random number generator for efficient simulations.
• Provision of standard inference and prediction routines, with direct simulator access for custom needs.

Main Results:

• The packages automate the generation of high-performance C++ code from domain-specific languages, reducing manual effort.
• Guaranteed reproducibility and performance enable researchers to focus on model conceptualization.
• Successful application in real-time COVID-19 pandemic modeling efforts.

Conclusions:

• The developed R packages significantly streamline state space model design and deployment.
• These tools enhance efficiency, reproducibility, and performance in complex modeling tasks.
• The packages offer a valuable resource for infectious disease modeling and other scientific fields requiring sophisticated simulations.