A parallel implementation of spatially distributed stochastic chemical kinetics

Stochastic simulation of reaction kinetics has emerged as animportant computational tool in molecular systems biology and will likely continue to grow in importance as experimental techniques are further developed and spatial models can be calibrated to biological data. Many applications require a large number of sample realizations to be generated in order to allow for useful statistical analysis.Exploring different parameters or estimating responses to stimuli adds further complexity such that the generation of tens of thousands of independent realizations is not uncommon. For these applications computational efficiency is an important concern. This study concerns a replica parallel implementation of a stochastic simulationalgorithm with the aim of increasing simulation efficiency and explores related concepts such as random number generation in the context of multiprocessing. A handful of different random number generators were profiled in a multiprocessing context. This led to the identification of performance issues with the drand48 random number generation algorithm and allowed for careful selection of a more appropriate random number generator to be used with the parallel implementation. Two alternative solutions were implemented and integrated into the URDME framework. Results showed that the parallel implementations reached speedups up to around 3x on a 12core machine compared to the serial solution when simulating multiple replicas