Mathematical Biology Seminar

Time-Dependent Parameter Estimation for Infectious Disease Models

-
Speaker(s): Susan Rogowski (North Carolina State University, Mathematics)
Due to the high dimensionality of infectious disease models and the presence of observational noise and reporting errors in real-time data, parameter estimation is an ongoing challenge for these applications. The added complexity of time-dependent parameters amplifies the challenge of accurately calibrating these models. The recent COVID-19 pandemic highlighted these issues and the need to develop more robust and efficient time-dependent parameter estimation methods. In this talk, we present two methodologies for calibrating epidemiological models. The first is a modified version of a regularized predictor-corrector algorithm aimed at stable low-cost estimation of infectious disease parameters. This method is applied to a novel compartmental disease model describing COVID-19 dynamics, which accounts for vaccination and immunity loss. We apply this algorithm to infer the time-dependent transmission rate and the effective reproduction number for a COVID-19 compartmental disease model. The second method is an equation learning method designed to learn and calibrate a compartmental disease model from agent based model data. This is done through using a Biologically Informed Neural Network (BINN). This second method builds a framework that exploits the advantages of agent based and compartmental models in order to better inform policy decisions for infectious disease outbreaks.

Physics 119