STEPS4 simulation models for submission to Frontiers in Neuroinformatics

Introduction

In this repository we present the models described in the manuscript submitted to Frontiers in Neuroinformatics, titled STEPS 4.0: Fast and memory-efficient molecular simulations of neurons at the nanoscale.

Please note that the solution described in the manuscript focusing on large scale cpu-based high performance computing clusters, thus some of the models may not be suitable for regular desktop machines. We also provide the data we gathered in our study for readers who are interested in statistical analysis of the existing results.

In order to run the models an installation of STEPS 4.0 is required. Please clone and build STEPS 4.0 from: https://github.com/CNS-OIST/STEPS.

Installation of STEPS

Without loss of generality, let us assume that you installed STEPS in /path/to/STEPS, see STEPS installation documentation for more details. In order to make the simulator discoverable by Python, it is required to provide its location using the following Bash command:

export PYTHONPATH="/path/to/STEPS:$PYTHONPATH"

After, you are ready to run the models. Additionally, you can:

export STEPS_INSTALL_DIR=/path/to/STEPS

and, use it to locate the installed STEPS.

Running the models

For more information on how to run a specific model we suggest to check the specific README.md in the model folder. In every case, at the end a file res*.txt is generated in the respective results folder. It contains the raw traces that can be statistically analyzed.

Validations

To analyse the raw traces we suggest to download https://github.com/CNS-OIST/STEPS_Validation and head toward the folder postproc. In there you can copy (in the appropriate subfolder) the traces and perform some statistical analyses. Check the relative README and the example for more information.

The complete validation dataset for the Frontiers publication is available at https://doi.org/10.5281/zenodo.7194567.

Profiling/ Performance

The measurement of STEPS performance is achieved through suitable instrumentation. To simply generate all the graphs found in the submitted manuscript, locate the STEPS_PerfGraphs.ipynb jupyter notebook in the ./profiling folder, and run all the cells of the notebook. All data needed is already baked in the notebook. However, the interested reader/scientist could follow the instructions found in ./profiling/README.md and perform the strong scaling & Roofline analysis to re-generate similar data, using for example a different supercomputing facility.