[Neuroinfo] Public release of STEPS 3.0, a parallel reaction-diffusion simulator.

[Iain Hepburn]

The STEPS team is pleased to announce the release of STEPS 3.0. The software is available through a public Github repository (https://github.com/CNS-OIST/STEPS_Release) and supporting documentation and other information can be found via the STEPS homepage (http://steps.sourceforge.net/).

STEPS is designed primarily to simulate molecular neuronal pathways with stochastic interactions within complex neuronal geometry described by tetrahedral meshes. Since STEPS 2.0 the membrane potential can be solved on a surface of the tetrahedral mesh allowing coupling between the biochemical pathways and electrophysiological activity of the cell, which was first demonstrated in a stochastic model of Purkinje cell calcium bursts (http://www.jneurosci.org/content/33/40/15848.short). 

These initial applications demonstrated the limits of serial applications with long runtimes in large-scale models, calling for the need to make use of parallel architectures. In two recent papers we describe how parallelization of STEPS was achieved, first with a theoretical paper describing how the original serial algorithm was adapted to a more scalable operator-splitting approximation with minimal loss of accuracy (http://aip.scitation.org/doi/full/10.1063/1.4960034) and then describe initial applications of MPI-based parallel STEPS on High Performance Computers (http://journal.frontiersin.org/article/10.3389/fninf.2017.00013/full). We found that parallel STEPS gave a maximum speedup of 500 times on 1000 cores for the reaction-diffusion component of the calcium burst model and a speedup of 3500 times on 2000 cores for a more simplified version of the model.

STEPS 3.0 is the first release of STEPS under collaboration between the Okinawa Institute of Science and Technology and the Blue Brain Project and contains the first MPI-based parallel solver including an initial parallel ‘EField’ membrane potential solution, as well as other changes and additions such as vastly reduced mesh-loading times and an improved build system based on CMake.

The STEPS team would be happy to answer any queries, either through the STEPS Support group (https://groups.google.com/forum/#!forum/steps_support) or via email (steps.dev at gmail.com). 

Best regards, 
The STEPS team