Community structure and epidemic prediction
Peter Dawson – Centre for Complexity Science - University of Warwick
Typically limited resources are available in the event of a foreign incursion of a livestock disease.
Effectively targeting these resources in the early stages of an outbreak is of great importance.
Community detection is a network analysis tool that searches for modular structure in a network. We
look at the network of cattle farms1 in Great Britain and the same network aggregated at a county-level
and analyse their community structures in the context of an epidemic.
Fig 1: Farm-level community detection with
a Newman-Girvan null model2
Fig 2: County-level community detection
with the Newman-Girvan null model2
Disease modelling on a dynamic network.
Infectious diseases can be easily modelled on a
network using the SIR model. Nodes can be in
either Susceptible to infection, Infectious or
Recovered. Our model uses farms as nodes and
edges are cattle movements between farms.
A farms is initially infected, it
may infect other farms if
there are cattle movements
between them.
Fig 3: County-level community detection
with a spatial null model3
Results – comparing community structures.
We seeded an infection and ran 100 simulations
for every farm in Great Britain. We then
compared the number of infected farms in the
seed farms community IC with the total number
of infected farms I.
0.5
Counties
Counties
County−level
NG
Farm−level
NGlevel NG
County
County−level Spatial
0.45
Farm level NF
County level spatial
0.4
0.35
0.3
0.25
0.2
The network is dynamic,
edges change every day
but frequent edges may
persist.
Farms eventually recover
and can not be re-infected.
Infection is determined by
the number of infectious
farms, where as recovery is
an internal process.
0.15
0.1
0.05
0
1
2
3
4
5
6
7
8
9
10
Ratio of IC /I per decile
Fig 4: Histogram of the ratio of IC/I per decile for the three community
detection methods and compared to using defined counties
1Data
is from the 2010 Cattle Tracing System
supplied by Defra, 2Newman, Girvan Phys Rev
E 2004, 3Expert et al. PNAS 2010