Stochastic simulator for a complex biological system

Condor is used intensively by several research projects in this School

Prof. Richard Boys writes

"One of our PhD students is using Condor as part of her research to successfully speed up computationally intensive tasks which would otherwise take an impractically long time. A particular example is outlined below.

She has built a stochastic simulator (computer model) for a complex biological system. The system depends on many input parameters and, as the simulator is stochastic, running it at the same input values gives different outputs each time, essentially producing an output distribution for each input combination. Since the system is large, each simulator run is slow, taking in the region of an hour.

The research problem is to determine values of the inputs which best explain a particular set of experimental data. In order to construct an inference engine to do this, the simulator needs to be run many millions of times. This is clearly not feasible. Instead we build a stochastic approximation to the simulator (called an emulator) which is used instead of the simulator in the inference engine.

To construct the emulator, we still need to run the simulator at a large number of input combinations, of the the order of hundreds or thousands. As each simulator run is independent, this process is ideal for parallelisation.

She found it fairly straightforward to modify the C code developed for this problem to work with Condor, and typically runs around 250 - 300 jobs at once using Condor with little additional computational overhead. On average a simulator run takes around an hour, and so running 500 jobs on one machine would take roughly 3 weeks. However, using Condor this task can be done in only a few hours.

At the moment, many input parameters in her simulator model are fixed. However, the next stage of the research project is to decrease the number of fixed parameters in order to better explain and understand the experimental data. With this in mind, it would be very useful to have more fast computers running LINUX within the Condor system."