NEST Desktop, an Educational Application for Neuroscience

Workflow in client-server architecture and technical challenges. A, Simplified relationship of the client (front end) and the server (back end). B, The front end handles network construction and data analysis, whereas the back end executes simulations with NEST (Gewaltig and Diesmann, 2007). C, Problem formulations regarding the software architecture (left column) together with the technical solutions implemented in NEST Desktop (center column) and associated examples (right column).

Software development and documentation of NEST Desktop. The code is open source and available on GitHub. The development of NEST Desktop makes use of PyPI and Docker Hub. An associated docker image containing NEST Desktop together with the simulator NEST can be pulled on any local machine or server infrastructure. The ReadTheDocs platform provides detailed information about NEST Desktop for users, lecturers, deployers, and developers.

Network editor, activity explorer, and lab book share a common user interface. The steps of NEST Desktop’s conceptual approach for rapid prototyping (right; adapted from Fig. 1) correspond to the distinct views of the tool. A, The network editor provides interactive construction and modification of the neuronal network. B, The activity explorer visualizes network activity. C, The lab book gives users a complete picture of a constructed network. The project manager is shown on the left.

Visual network construction. A, The network editor is the main work space to graphically construct a network graph (left) and adjust network properties with the controller panel (right). Stacks of node and connection parameters associated with the chosen models are displayed in colored panels. B, A right click with the mouse in the blank space of the editor opens a creation panel (left) to add a node to the network. Node types are stimulator (S), neuron (N), and recorder (R). Connections between nodes are drawn with the mouse cursor (top right). A minimal network may consist of a stimulator, a neuron, and a recorder (bottom right). C, The network controller to the right of the network graph allows users to select and parameterize models. Clicking once on the model name (top left) opens a popup for selecting parameters via checkboxes (middle); clicking there twice allows the user to select a different model from a drop-down menu (bottom left). When a subset of model parameters is selected, the corresponding values can be modified (right) by moving sliders, incrementally increasing and decreasing the value, or by directly entering the value. A restart of NEST Desktop is also possible.

Code generation through visual network construction. The graphically composed network (Fig. 5) is automatically translated into textual source code. Visual elements in the network graph (shapes for nodes and arrows for connections) are associated with generated code lines. The resulting script is a complete definition of a simulation experiment in PyNEST with code blocks to be executed in succession: Create nodes, Connect nodes, Start simulation, and Get activity. The sketched network of only three connected nodes (stimulator to neuron to recorder) is a minimal example for illustration; further details such as parameter values set via the GUI are also turned into code.

Data analysis and visualization. The NEST simulator executes the network simulation and returns recorded activity data to be analyzed and visualized (top left). Quasi-analog signals like membrane potentials and discrete spike times can be displayed across time (bottom left). Such visualization is accompanied by basic analysis like the computation of spike counts. If the neurons in the simulated network are arranged in space, a 2D or 3D animation offers a view of the ongoing activity at the respective neuronal positions (bottom right). Calculated quantities are presented in table format (top right).