Research Interests
Direct Links
Relevant Links


Tutorial 02 - Susceptible-Infected model

This tutorial demonstrates the simplest possible dynamic process simulating spread of epidemics in a population. The contagion process, Susceptible-Infected (SI), is frequently used to model spread of contageous diseases or information in societies. Here we use the Complex Networks Toolbox to simulate diffusion on regular network (square grid) and random scale free network. The script running the tutorial can be dowloaded from here.

The key assumptions are:

  • Each node can assume one of two states: S for susceptible and I for infected.
  • Initially, all but few nodes are in state S.
  • The infection spreads along the network ties, from infected nodes (I) to its susceptioble (S) neighbors.
  • Once infected, the node remains infected for ever.
  • β represents the probability of the infection to spread along the tie per simulation step.
  • Multiple exposures are independent. In other words, the probability of a node having k infected peers to switch to the infected state is (1-(1-β)^k) per unit time.
  • We will assume for simplicity that a node cabnnot spontaniously switch states (although, this is regulated by a single parameter).

Few more remarks before we start

  • Eventually all (connected) nodes will be infected for any non-zero value of β.
  • There is a number of analytical solutions (i.e. based on mean field assumption) to SI and other related models. Barrat, Alain, Marc Barthelemy, and Alessandro Vespignani (2008), Dynamical processes on complex networks: Cambridge University Press. p. 184 is a good place to start further reading.

Let's start with a plain square lattice.
Generate the lattice, initialize the dynamic process and make sure that at least one node is infected

m = 100; % lattice dimentions
n = 100; PInitialInfection = 3/10000; %the probability of infection at start. On average we'll start with 3 nodes PSpontaneousInfection = 0; % the probability of spontaneous infection (not due to contagion) as the simulation runs.
PTransfer = 5/100; % the probability of infection transfer per link per unit time.
Graph = GraphGenerateSquareLattice(m,n); % create regular graph
Degrees = GraphCountNodesDegree(Graph); % compute degrees.
DP_SI = DynamicProcessInitializeSI(Graph,[],PInitialInfection,PSpontaneousInfection,PTransfer);
% initialize SI for the lattice. % Second parameter ([]) instructs the framework to extract node IDs from the Graph (i.e. no isolates).
% third parameter - probability of infection at start. On average we'll start with 3 nodes
% next - the probability of spontaneous infection (not due to contagion). as the simulation runs.
% last parameter - the probability of infection transfer per link per unit time.

while nnz(DP_SI.States)==0
% ensure that at least one node is infected at startup. Otherwise, infection will not start (spontaneous adoption,PSpontaneousInfection, is set to zero).
DP_SI = DynamicProcessInitializeSI(Graph,[],PInitialInfection0,PSpontaneousInfection,PTransfer);
Prepare the fugure
figure; h1 = subplot(2,1,1); % top panel is going to show the total number of infected nodes
h2 = subplot(2,2,3); % bottom-left panel showing the number of nodes infected at each time step
h3 = subplot(2,2,4); % shos the average degree of the newly infected nodes.
AverageDegree = mean(Degrees(DP_SI.UserData.TimeLine==DP_SI.Time,3)); % compute the average degree of the initially infected nodes.
Run the model
while any(DP_SI.States~=true) % continue the simulatiion as long as there's at least one node
DP_SI = DynamicProcessIterate(DP_SI,1); % perform 1 simulation step
% plot the results
[NumberOfInfected, TimeAxis]= hist(double(DP_SI.UserData.TimeLine(DP_SI.UserData.TimeLine~=-1)),0: DP_SI.Time); % compute the number of infected at each time step.
plot(TimeAxis, cumsum(NumberOfInfected),'*:b'); xlabel('Time'); ylabel('Cumulative number of infected'); title(sprintf('time=%d',DP_SI.Time)); % plot cumulated infected
subplot(h2); plot(TimeAxis, NumberOfInfected,'*:r'); xlabel('Time'); ylabel('Number of infected'); % plot the number of infected per time step over time
if any(DP_SI.UserData.TimeLine==DP_SI.Time) % compute and plot the average degree of the recently infected node.
AverageDegree(end+1) = mean(Degrees(DP_SI.UserData.TimeLine==DP_SI.Time,3));
AverageDegree(end+1) = NaN; %average degree doesn't exist if there are no new infected nodes.
subplot(h3); plot(TimeAxis, AverageDegree,'*:r'); xlabel('Time'); ylabel('Average degree');
drawnow % update the plot
The following figure shows the number of infected nodes over time (top), the change in the infection rate (bottom-left) and the average degree of the infected nodes over time (bottom-right). Let's generate a video of the infection diffusing on the square lattice
AviObj = VideoWriter('SI.avi');
AviObj.FrameRate = 10;
AviObj.Quality = 100;
FreshnessRange = 15; % nodes infected within this interval of time are still considered "fresh" and colored as such
for i =0 : double(max(DP_SI.UserData.TimeLine))+FreshnessRange
   Map = zeros(m,n);
   Map(DP_SI.UserData.TimeLine~=-1 & DP_SI.UserData.TimeLine<(i-FreshnessRange+1)) = 1;
   Map(DP_SI.UserData.TimeLine~=-1 & DP_SI.UserData.TimeLine<=i & DP_SI.UserData.TimeLine>=i-FreshnessRange+1) = 2;
   Map(Map==0) = NaN;
   Map(1,1) = 2;
   pcolor(1:n, 1:m, Map);
   shading interp
   CurrentFrame = getframe;

Same video can be downloaded from here (41mb).

Now, let's run the same simulation on a scales-free graph of the same size (10,000 nodes) and same average degree (~4.0) as in the square lattice in the example above.

Graph = mexGraphCreateRandomGraph(m*n,[1 : 200],[1 : 200].^-2.35); % create a graph of m*n nodes with power law distribution of degrees.
% The average degree of that graph is approximately 4, same as square lattice.
Graph = GraphMakeUndirected(Graph ); % make the graph undirected.
% the following code is identical to the Lattice case.
Degrees = GraphCountNodesDegree(Graph); % compute node degrees.
DP_SI = DynamicProcessInitializeSI(Graph,[],PInitialInfection,PSpontaneousInfection,PTransfer);
while nnz(DP_SI.States)==0
   DP_SI = DynamicProcessInitializeSI(Graph,[],PInitialInfection,PSpontaneousInfection,PTransfer);
h1 = subplot(2,1,1);
h2 = subplot(2,2,3);
h3 = subplot(2,2,4);
AverageDegree = mean(Degrees(DP_SI.UserData.TimeLine==DP_SI.Time,3));
while any(DP_SI.States~=true)
   DP_SI = DynamicProcessIterate(DP_SI,1);
   [NumberOfInfected, TimeAxis]= hist(double(DP_SI.UserData.TimeLine(DP_SI.UserData.TimeLine~=-1)),0: DP_SI.Time);     subplot(h1);
    plot(TimeAxis, cumsum(NumberOfInfected),'*:b');
   xlabel('Time'); ylabel('Cumulative number of infected'); title(sprintf('time=%d',DP_SI.Time));
   plot(TimeAxis, NumberOfInfected,'*:r');
   xlabel('Time'); ylabel('Number of infected');
   if any(DP_SI.UserData.TimeLine==DP_SI.Time)
      AverageDegree(end+1) = mean(Degrees(DP_SI.UserData.TimeLine==DP_SI.Time,3));
      AverageDegree(end+1) = NaN;
   plot(TimeAxis, AverageDegree,'*g');
   hold on;
   plot(TimeAxis, ones(size(TimeAxis))*mean(Degrees(:,3)),'g-','LineWidth',4);
   hold off;
   ylabel('Average degree');

Although, this network is similar in size and the number of links to the lattice above, the infection spreads about 10 times faster (top plot). In fact, the bulk of the nodes get infected at about 50 times steps (as opposed to ~500 time steps in the case of lattice. This is caused by heterogeneity of node degrees. The highly connected nodes are more exposed to the epidemics than the average node and get infected very early. Each of the hubs in turn affects a bunch of other nodes (the graph in undirected) accelerating the spread.