sarscov2_unobserved

This repository contains code to reproduce results from the following publication

TA Perkins^*, SM Cavany^*, SM Moore^*, RJ Oidtman, A Lerch, M Poterek (2020) Estimating unobserved SARS-CoV-2 infections in the United States. PNAS doi:10.1073/pnas.2005476117

License

This code is being released under the MIT License.

Overview

Contents of this repository are organized according to code, data, results, and plots. All plots and results shown in the main text and supplementary material are generated in script_main.R and script_supplement.R respectively, with the exception of the sensitivity analysis. Folders for code, data, results, and plots related to the sensitivity analysis described in the supplemental text are located in a sensitivity subfolder within each of those folders. All code for the analysis featured in the main text was written in the R language (version 3.5.2) and was executed on a Lenovo Thinkpad running Linux (PureOS 9.0). The sensitivity analyses were written in R (version 3.6.2) and run on a Linux cluster. At the time the research was done, all R packages used in this analysis were available on CRAN and were straightforward to install and load.

Baseline analysis

To re-generate the figures and quantities related to our baseline analysis in the main text of the manuscript, simply run script_main.R. The model itself can take some time to run. To re-generate results without re-running the model, skip lines 258-299 and load saved model output in line 302.

To generate results with an alternative parameterization, do the following:

1. Input new parameterization in data/sensitivity/covid_param_estimates.csv and data/parameters.csv. To run just a single parameterization (i.e. no sensitivity analyses), enter just a single row in data/sensitivity/covid_param_estimates.csv.
2. Run steps 1 and 2 of the sensitivity analysis sections. This generates the posterior distributions for the importation detection probability and relative infectiousness of asymptomatic individuals.
3. Run script_main.R.

Sensitivity analysis

The results and plots for the sensitivity analysis can be recreated by running the following scripts in the code/sensitivity subfolder:

1. script_crc_estimate_parameter_sensitivity.R: This script runs a sweep of the AsympRfraction and parameters for each of 18 different parameter scenarios and calculate log likelihoods.
   1. Input: scenario run number (referred to as XX in input/output filenames)
   2. Output: mle_local_XX.rda object containing likelihood values
2. parameter_profiles_sensitivity_analysis.R: This script uses the log likelihoods from each parameter sweep, uses bic.grid to create a modified log likelihood surface, and then samples from that surface to get posteriors for the two estimated parameters.
   1. Input: mle_local_XX.rda
   2. Output: parameter_estimates_posterior_XX.rda
3. bic_profile_plots.R: Uses output files from step 2 to create hexbin plots of joint estimated parameter posteriors
   1. Input: parameter_estimates_posterior_XX.rda
4. Script_crc_params_sensitivity_fit. R: Script using the parameter posteriors to simulate infections, deaths, etc for each of 18 scenarios
   1. Input: parameter_estimates_posterior_XX.rda
   2. Output: sensitivity_sims_XX.rda
5. combine_sim_results_fit_import.R: Creates summary statistics for cumulative infections and plots for all 18 parameter sensitivity scenarios
   1. Input: sensitivity_sims_XX.rda
6. combine_sim_results_ratio_deaths.R: Creates summary statistics and plots of deaths compared expected after versus before March 13 for all 18 parameter sensitivity scenarios
   1. Input: sensitivity_sims_XX.rda
7. multiple_pDetect_ts.R: Creates plots of the probability of detecting a local symptomatic infection over time for all 18 parameter sensitivity scenarios
   1. Input: sensitivity_sims_XX.rda