Jingxue (Grace) Feng

Title: Advanced Bayesian Modeling for Public Health
Date: Tuesday, April 15th, 2025
Time: 10:00am
Location: Zoom
Supervised by: Dr. Liangliang Wang

Abstract:

Bayesian approaches are increasingly important in public health because they provide a formal framework for quantifying uncertainty within models, offering more informed and cautious estimates for decision-making or providing insights. This thesis explores Bayesian models to address key challenges in public health, including epidemic tracking while assessing external interventions and genomic data analysis for viral mutation clustering. This thesis starts with introducing Bayesian inference for state-space models, illustrating fundamental concepts and computational techniques such as Markov Chain Monte Carlo (MCMC), Sequential Monte Carlo (SMC), and Particle Markov Chain Monte Carlo (pMCMC). This framework is then extended to switching state-space models, which accommodate regime changes in dynamic processes. Both frequentist and Bayesian approaches are reviewed, with an emphasis on Bayesian inference techniques that enhance flexibility and interpretability.

To assess the impact of interventions on disease spread, we develop a Beta-Dirichlet switching state-space transmission model that simultaneously tracks underlying disease dynamics and intervention effectiveness. Implemented using pMCMC algorithm, this model efficiently estimates the temporal evolution of latent states and high-dimensional parameters. Applying the model to British Columbia’s COVID-19 outbreak, we quantify reductions in transmission rates following public health interventions. Furthermore, we introduce an innovative data analysis pipeline that integrates genomic and clinical data to investigate viral mutation clustering based on their associations with clinical features. Using model-based clustering techniques, the pipeline organizes mutations according to their shared associations with these features. Bayesian inference is performed using an adaptive MCMC approach to precisely estimate high-dimensional parameters. Applying the model to global COVID-19 data reveals four distinct clusters of SARS-CoV-2 mutations, each showing unique associations with selected clinical features.

Keywords: Bayesian models; public health; switching state-space model; Sequential Monte Carlo; particle Markov Chain Monte Carlo; model-based clustering; adaptive Metropolis-Hastings within Gibbs sampling