StochPy: Stochastic modelling in Python
StochPy (Stochastic modelling in Python) is an easy-to-use package, which provides several stochastic simulation algorithms (SSAs), which can be used to simulate biochemical systems in a stochastic manner. Further, several unique and easy-to-use analysis techniques are provided by StochPy.
SSA’s try to describe the time evolution of a reacting system, such that it takes into account discreteness and stochasticity. Discreteness and stochasticity play often important roles in biochemical systems. Therefore, SSA’s are widely used to simulate biochemical systems. Especially for biochemical systems that contain low copy numbers. For such systems, deterministic models often fail to capture the stochasticity of the system, while SSAs are capable of capturing this stochastic behaviour.
In addition, StochPy can be used as a plug-in into PySCeS. PySCeS - the Python Simulator for Cellular Systems - is a toolkit for the analysis and investigation of cellular systems.
StochPy's features
Stochastic Test suite
StochPy has been successfully tested against the SBML stochastic test suite (Evans, T. W., Gillespie, C. S., Wilkinson, D. J. (2008) The SBML discrete stochastic models test suite, Bioinformatics, 24:285-286.). This test suite can be downloaded here. StochPy was tested for parameter overloading, usage of boundary conditions, mathematical expression parsing, assignments and timed events. Note that StochPy automatically converts species concentrations to species amounts.
One nice example about StochPy and SBML molecule number based events is given on the example webpage. All other results (1000 trajectories) can be found here. In addition, one can download this script to run the entire test suite for all implementations that are available in StochPy. 1000 trajectories are generated for each model in the test suite, except two extremely slow models. Generating 1000 trajectories of all models and calculating all interpolated results should take about 30-40 minutes per stochastic simulation method on a normal desktop. As mentioned in the paper of Evans et al., the number of trajectories should not be less than 1000. However, for the statistical tests to have reasonable power to detect problems, the number of trajectories should be set to at least 10,000 (but this will be very time-consuming for some models).