Re: [Swarm-Modelling] modelling disease spread

[Marcus G. Daniels]

Hi Sam,
> Am planning to do some disease spread modelling n stuff using agent 
> based modelling.I have selected Foot and Mouth disease as platform of
> interest.
> Does ne1 know of any similar research conducted earlier and hints of 
> how to go
> abt would be much appreciated.
For a spatially explicit model where animals get the virus through feed, 
water, and other animals, you could start with a grid of interaction 
regions.
An interaction region would be big enough to model densities of animals 
where there was a risk of transmission, but small enough to keep 
geographically separate herds. Each region could have a set of animals 
in it.   For example, an animal could have an x,y coordinate as on a 
map, and there would be 10x10 or 100x100, etc. grid points in each 
region.   Figuring out what region the animal was in would simply 
involve integer division by that number of grid points on each 
dimension.   It would be the job of each animal agent to add or remove 
itself from the appropriate region as it moved around.   So if you had 
1000 animals and 100 regions, if everything was uniformly distributed, 
you'd expect to have 10 animals per region, and you could enumerate the 
animals in each because each region has a list that is growing and 
shrinking as animals move around.   And, because you can enumerate the 
animals in the regions, it is easy to implement disease spread:  you 
take the subset of animals in each region that have the virus and 
communicate with some probability it to others in the region.  (If there 
are other important reasons why animals in a region would need to 
interact, then the virus spread just becomes another facet of that 
interaction.) 

To measure the overall spread of disease you can also keep track of all 
of the animals in a single list and just count the number of the total 
living animals that have had the virus communicated to them.

Generalizing this things like stationary contaminated feeding sites 
becomes a matter of having another grid (in Swarm we'd call it a 
Discrete2d) that can be related through appropriate scale normalization 
of the animals x/y coordinate on each move event.  This way, not only 
can disease be communicated from iterating through the animals in a 
region, but now also by looking at another grid having unmoving risk sites.

You could start by writing code to populate the world in `plausible' 
ways (e.g. realistic herd densities), and then seek to get real data on 
real landscapes and known outbreak scenarios.  Much of the detail you 
add in a simulation will depend on what kind of field and validation 
data you can get. 

Marcus