Main content

Infectious disease epidemiology in the era of deep learning

Show full item record

Title: Infectious disease epidemiology in the era of deep learning
Author: Augusta, Carolyn
Department: Department of Mathematics and Statistics
Program: Mathematics and Statistics
Advisor: Taylor, GrahamDeardon, Rob
Abstract: In this thesis, we will explore deep learning and infectious disease modelling. In chapter 2, we use deep learning methods to match an epidemic curve to its generating model. We simulate two populations of swine farms, one dense and one sparse. Then we simulate many epidemics over these two populations, recording the number of infectious farms at each time point. Finally, we train classification models to determine the most likely epidemic simulation model that generated the observed counts. We believe we are the first to apply LSTMs and GRUs in an epidemic classification context. After this exploration, we move on in chapter 3 to consider the potential for contact-based disease spread in a real population. We analyse a proprietary data set on shipping among swine farms in Manitoba, Canada. The epidemiologically relevant measures we provide of the dynamic shipping network may be used to simulate disease spread. Although the swine industry contributes over a billion dollars to the GDP of Manitoba (Government of Manitoba, 2019b), we are the first to publish an exploration of the dynamics of swine shipment in Manitoba with a view to inform epidemic mitigation efforts. Then, in chapter 4, we extend current methods of dynamic link prediction to allow a richer representation to drive predicted graph dynamics. That is, the goal is to predict which vertices will have an edge between them at a future time point, based on the past history of a dynamic graph. By learning how and when edges change, we can predict how an infectious disease may spread in the population. We demonstrate the superior performance of our method on two publicly available data sets. Finally, we investigate the impact of various physical distancing scenarios in the province of Saskatchewan during COVID-19. Saskatchewan was the rst province to announce a reopening plan. We found that although reopening the province's economy is of great importance, to do so prematurely could cause a dramatic increase in cases and recurrent epidemic peaks. Altogether, this thesis discusses deep learning and other model types applicable to infectious disease epidemiology.
URI: https://hdl.handle.net/10214/21317
Date: 2020
Rights: Attribution-NonCommercial-NoDerivatives 4.0 International
Terms of Use: All items in the Atrium are protected by copyright with all rights reserved unless otherwise indicated.
Related Publications: Augusta, C., Deardon, R., and Taylor, G.W. (2018). Deep Learning for Supervised Classification of Spatial Epidemics. Spatial and Spatio-temporal Epidemiology. 29. 187-198. https://doi.org/10.1016/j.sste.2018.08.002Augusta, C., Taylor, G.W., Deardon R. (2019). Dynamic contact networks of swine movement in Manitoba, Canada: Characterization and implications for infectious disease spread. Transboundary and Emerging Diseases. 66(5). 1910-1919. https://doi.org/10.1111/tbed.13220


Files in this item

Files Size Format View
Augusta_Carolyn_202009_PhD.pdf 6.250Mb PDF View/Open

This item appears in the following Collection(s)

Show full item record

Attribution-NonCommercial-NoDerivatives 4.0 International Except where otherwise noted, this item's license is described as Attribution-NonCommercial-NoDerivatives 4.0 International