Data science is a field of growing interest
amongst both the
public and scientific communities. However, data science methodology
does not use the insights of dynamical systems theory as much as it
could, compared to widespread applications of conventional statistics.
In this talk I will describe an application of dynamical systems theory
to a data science problem. In particular, vaccine scares are of great
concern to population health, because they can enable renewed infectious
disease outbreaks and delay global eradication by many years. Vaccine
scares often entail coupled dynamics between social vaccinating
behaviour and disease transmission dynamics that can be captured by
simple systems of nonlinear differential equations. These equations
exhibit bifurcations that are often termed “critical transitions”, where
the state of the system shifts abruptly to a contrasting state as a
parameter is moved beyond a bifurcation point. While apparently
occurring without warning, in stochastic systems these transitions are
often preceded by an increase in time series autocorrelation and
variance prior to the transition, caused by the dominant eigenvalue
approaching zero. Therefore, it is possible that critical transitions
may be predicted ahead of time by such early warning signals. If
vaccine scares can be modelled as critical transitions, then we may be
able to predict them by looking for early warning signals. In this talk
I will describe and characterize some theory for critical transitions
and early warning signals in coupled behaviour-disease systems. I will
also present analyses of data during the 2014-15 Disneyland, California
measles outbreak. The data consist of time series of measles-related
Google searches, and measles-related tweets that have been
sentiment-classified into pro- and anti-vaccine tweets using
machine-learning algorithms. The data reveal the telltale signatures of
early warning signals before the 2014-15 Disneyland, California
outbreaks. Such methods may improve the ability of health authorities
to anticipate growing vaccine refusal, and focus messaging strategies
accordingly. We suggest that data science can benefit from greater
interaction with dynamical systems theory.
Chris Bauch is a biomathematician in the
Department of Applied
Mathematics at the University of Waterloo. He studies epidemiological
and ecological systems with a particular emphasis on evaluating
interventions, and coupling models of human behaviour with models of
disease dynamics or ecological dynamics. Approaches include
differential equations, stochastic simulations, and network models. He
has published over 80 papers in journals including Science, PNAS,
Proceedings of the Royal Society B, PLoS Computational Biology, Lancet
Infectious Diseases, and others. His research has been funded by NSERC,
CIHR, GlaxoSmithKline, the United States Food and Drug Administration,
The Ministry of Health of Ontario, and the World Health Organization.
Some of this work has reached a wide audience through the media, and has
been written up in The New York Times, Scientific American, USA Today,
BBC News and other media.
Contact at the MS2Discovery Research Institute:
Manuele Santoprete (Host of the speaker,
Multidisciplinary Talk, Tectons 3, and others)
Refreshments will be provided