Chalmers Conferences, 9th European Conference on Mathematical and Theoretical Biology

Geometric Epidemics: An application of differential geometric sampling methods to stochastic epidemic models
Samuel Charles Bilson

Last modified: 2014-03-31


Infectious diseases are a major cause of mortality and morbidity worldwide. In recent years, epidemiological models have played an important role in understanding disease spread and informing control policy. Such models, as in most of mathematical biology, are non-linear and can be highly complex with multiple parameters. Thus precise estimation of parameters to given datasets, whilst controlling for model complexity and computational efficiency, is a vibrant area of current research. Traditionally, approaches to parameter estimation have chosen to either sacrifice likelihood-based model fitting or biological understanding for practical purposes. One such problem has been the application of MCMC sampling methods to model posteriors where certain combinations of parameters are insensitive to large changes in model dynamics, a so-called 'sloppy' problem. However, by considering a differential geometric sampling method (RMHMC), one can impute more principled model fitting techniques, with accurate parameter and uncertainty quantification to the inverse problems which arise in biology. In this talk we describe the application of RMHMC to parameter estimation of fitting a general class of compartmental epidemic stochastic SEIR models to shedding data from the recent H1N1 pandemic. We show that the likelihoods can be derived analytically, and we compare different sampling techniques for computational efficiency.


Epidemic; Markov Chain Monte Carlo; Riemann manifold sampling methods; parameter estimation