Bayesian Inference for Stochastic Epidemic Models via the Linear Noise Approximation
The stemr R package implements a Bayesian data augmentation algorithm for fitting stochastic epidemic models (SEMs) with arbitrary dynamics to partially observed epidemic count data via the linear noise approximation (LNA). This repository and package exist to expose and archive the code used in the paper, “A Linear Noise Approximation for Stochastic Epidemic Models Fit to Partially Observed Incidence Counts” by J. Fintzi, J. Wakefield, and V.N. Minin. The package provides facilities for exact simulation of SEMs via Gillespie’s direct algorithm, and for simulation of approximate epidemic paths via the LNA. Inference is carried out using a Bayesian data augmentation algorithm. LNA paths are sampled via the elliptical slice sampling algorithm of Murray et al. (2010), and the package offers several choices for adaptive and fixed kernel MCMC algorithms for parameter updates. The package also allows for time varying covariates, time varying parameters, and deterministic forcings. These functionalities are not yet fully documented, but sample code is available in the ‘/.development_files’ directory on the GitHub repository.
Package installation
To install the stemr package, clone this repository and build the package from sources. You should be able to rebuild in the usual way once you clone the package repo and install the other dependencies (odeintr, extraDistr, Rcpp, RcppArmadillo, and BH). Computationally intensive components in stemr package are implemented in C++. Hence, it is important to also make sure that your C++ toolchain is set up properly, e.g., by following instructions given in the Stan documentation, and that Rtools has been added to your system path. If you are working on a Windows machine, you may need to take additional steps to ensure your toolset is in order.
Vignettes and Simulation Code
The code for reproducing the simulations in Fintzi, et. al (2020) can be found in the ‘/.development_files/incidence_lna_simulations’ folder. There are two vignettes included in this package to help familiarize users with its basic functionality and that reproduce models the SEMs fit via the LNA in Fintzi, et al. (2020). The stemr vignette provides an introduction to the stemr package, and demonstrates how to simulate from and fit an SIR model via the LNA and ODE. The ebola_westafrica vignette demonstrates how to simulate from and fit a multi-country model for Ebola transmission, and then provides code to fit the model to data from the 2014-2015 outbreak in West Africa.