Assisting the mitigation of emerging infectious diseases using artificial intelligence (OZR3863)

Emerging infectious diseases have a significant impact on public health and
global economies. Controlling such epidemics is challenging, and in such a
complex environment, potential mitigation strategies need to be evaluated
using geospatial epidemiological models. The limited availability of case reports
during emerging epidemics complicates the fitting of such models.
Furthermore, the uncertainty about the progress of emerging epidemics
convolutes the decision-making process of policy makers.
Our goal is to develop a new real-time method to optimize the mitigation of
emerging epidemics of viral pathogens, for which we identify four objectives.
(1) Develop a methodology to use virus genomes as an additional
epidemiological marker for the estimation of parameters in a geospatial
epidemiological model. (2) Devise a method to automatically learn optimal
geospatial mitigation strategies, considering the current state of the epidemic,
given by the models obtained in the first objective. (3) Construct a pipeline that,
for a given outbreak, continuously updates the distribution over
epidemiological models, as epidemiological markers arrive. Using this
distribution, we will learn mitigation policies using model-based reinforcement
learning, to advise policy makers. (4) Evaluate our method on recent epidemics
Research Council Regulations | Chapter 5, Article 19 – Centraal Werkingsreglement Onderzoek | approved AB 17.02.2020
Research proposal description clearly add expected outcome; use up to 1500 words
Emerging infectious diseases have a significant impact on public health and global economies
[Jones,2008]. Over the past decade, several emerging viral pathogens caused epidemics across
the globe. The list of epidemics includes the Ebola virus that keeps sparking epidemics in Africa
[Dudas,2017], the recent yellow fever epidemic in Brazil [Faria,2018] and the ongoing SARSCoV-2 pandemic [Davis,2021]. Furthermore, due to climate change, we expect the migration
of tropical pathogens towards territories formerly devoid of transmission, causing a new wave
of public health emergencies [Altizer,2013].
The control of emerging epidemics is notoriously challenging, as there is typically a great deal
of uncertainty associated with the pathogen’s characteristics and the course that the epidemic
will take [Metcalf,2017]. In such a complex environment, potential mitigation strategies need
to be evaluated using reliable epidemiological models. However, modelling epidemics when
they emerge is difficult, as the number of infected individuals is typically limited at this phase
of the epidemic, which convolutes the fitting of such models [Britton,2019]. Additionally, most
mitigation strategies are to be implemented in an explicit geospatial context (e.g., vaccine
trials, insect reduction campaigns and border control). Therefore, epidemiological models
should be able to take this context into account, which further complicates the model and the
(i.e., Ebola, Yellow fever and SARS-CoV-2), which will lead to important insights
on how to respond to these and similar pathogens and future pandemics.
While we focus on epidemiological emergencies, this research will lead to new
general-purpose reinforcement learning algorithms to support decision makers
in real time.

Keywords:Reinforcement learning in complex models, Reinforcement learning under uncertainty, Realtime decision support with reinforcement learning, Epidemiological models, Bayesian phylodynamics

Disciplines:Machine learning and decision making, Adaptive agents and intelligent robotics, Epidemiology, Computational biomodelling and machine learning