Published on December 27th, 2013 | by Emily Corbett0
Approximated Bayesian Computation
By Lydia Braunack-Mayer, University of Adelaide
This student took part in the 2012/13 AMSI Vacation Research Scholarship program. For more information on this years program please click here
In 2009 swine flu (H1N1 influenza) hit Australia. People all across Australia were infected. Airport security was heightened, doctors were asked to report any case of swine flu they saw, and some schools were even shut down. Luckily for us, this strain of influenza was not too serious. Having swine flu was just like having a regular bout of influenza. Some people didn’t even know that they had it. Looking back, it seems a bit like a huge over-reaction.
Swine flu was not deadly, but the next influenza might be. So, what can we learn from the H1N1 influenza epidemic?
The H1N1 influenza virus spread throughout Australia extremely quickly; if a deadly influenza did arrive in Australia, we would need to act quickly in order to contain the disease. The 2009 pandemic gave us important information about the way that an epidemic in the future might spread. By looking at data from the swine flu epidemic we may be able to predict the spread of a more deadly infection. If we can do that, then we can act in order to contain have or maybe even prevent an epidemic.
In my AMSI vacation project I worked with a fellow AMSI vacation scholar, Brock Hermans, to predict the spread of infectious diseases. We approached this by finding a way to estimate the chance that one person with a disease might infect another. Estimating these infection rates is a relatively easy task if we have epidemic data that is simple to work with. Estimation is a matter of working directly with the data. However, the data isn’t always easy to work with. It can sometimes be very big and very complex. We wanted to find a method for cases when we can’t extract the information we need directly from the data.
We explored a method called approximated Bayesian computation, or ABC for short. ABC is a way of generating an imitation data set. This imitation data is a kind of simulated epidemic that is almost exactly the same as the real epidemic. Our simulated epidemic has the bonus of being in a form that is easy to work with. We can estimate the chance that once person might infect another in our simulated epidemic. Because our simulated epidemic looks almost exactly the same as the real epidemic, working out the transmission rates on our simulated data is the same as working out the transmission rates for the real epidemic. Our simulated epidemic has exactly the information we need in order to estimate the infection rates in the real epidemic.
Brock and I found a way to estimate the chance that, in an epidemic, one infectious person might infect another. With this we can predict the spread of a future outbreak. Hopefully work like ours will help to prevent an epidemic in the future.