BaySIR

BaySIR infers COVID-19 transmission dynamics using a Bayesian semiparametric probabilistic compartmental model to estimate time-varying epidemiological parameters and undocumented infections.

Key Features:

• Probabilistic Compartmental Model: Employs an extended susceptible-infectious-recovered (SIR) model that accounts for undocumented infections and allows epidemiological parameters to vary over time.
• Bayesian Inference Framework: Performs Bayesian inference by incorporating prior knowledge and updating with observed case data to quantify uncertainty in parameter estimates.
• Gaussian Process Prior for Transmission Rate: Uses a Gaussian process prior to model nonlinear, time-varying transmission rates without imposing specific parametric forms.
• Parallel-tempering Markov Chain Monte Carlo (MCMC) Algorithm: Implements a parallel-tempering MCMC algorithm to sample efficiently from highly correlated posterior distributions.
• Posterior Predictive Distributions for Forecasting: Generates posterior predictive distributions to produce forecasts of future observations under inferred model parameters.

Scientific Applications:

• Public health surveillance and planning: Provides detailed inference on transmission dynamics to inform public health officials and researchers.
• Intervention assessment: Assesses the effectiveness of interventions such as social distancing, lockdowns, and vaccination campaigns by estimating changes in transmission rates.
• Empirical and simulation validation: Applied to daily case data from Washington, New York, California, Florida, Texas, and Illinois and evaluated using simulated datasets.

Methodology:

Integrates daily confirmed COVID-19 case data into a state-space framework using an extended SIR compartmental model with time-varying parameters; applies a Gaussian process prior on the transmission rate; performs Bayesian parameter estimation via parallel-tempering MCMC; and samples posterior predictive distributions for forecasting.