ConGen-A Simulator-Agnostic Visual Language for Definition and Generation of Connectivity in Large and Multiscale Neural Networks

Abstract

An open challenge on the road to unraveling the brain's multilevel organization is establishing techniques to research connectivity and dynamics at different scales in time and space, as well as the links between them. This work focuses on the design of a framework that facilitates the generation of multiscale connectivity in large neural networks using a symbolic visual language capable of representing the model at different structural levels-ConGen. This symbolic language allows researchers to create and visually analyze the generated networks independently of the simulator to be used, since the visual model is translated into a simulator-independent language. The simplicity of the front end visual representation, together with the simulator independence provided by the back end translation, combine into a framework to enhance collaboration among scientists with expertise at different scales of abstraction and from different fields. On the basis of two use cases, we introduce the features and possibilities of our proposed visual language and associated workflow. We demonstrate that ConGen enables the creation, editing, and visualization of multiscale biological neural networks and provides a whole workflow to produce simulation scripts from the visual representation of the model.

Keywords: connectivity generation; connectome; large scale simulation; multiscale simulation; neural networks; visual language.

Figure 1

Hierarchical structure of a scene. Level 0 shows three superpopulations that group either populations or descendant superpopulations, as shown in level 1. Superpopulations SP_0_0 and SP_0_1 contain two neuron populations each, as depicted in level 2.

Figure 2

Creation of super-populations and populations. (A) The panel on the right sets the name root of the super-populations (SP in this example) and the number of entities (three in this example) to be created. (B) Right clicking on SP_2 allows the creation of descendant neuron populations. The panel on the right sets the name root of the populations (NP_2 in this example) and the number of entities (two in this example) to be created.

Figure 3

Creation of scenes in ConGen. (A) A scene where SP_0 has two hierarchical descendant levels (indicated by the two inner rings) while SP_1 and SP_2 have one hierarchical descendant level each. The green filling of the horizontal bars indicates that SP_1 and SP_2 have the same number of neurons while SP_0 has twice the amount. (B) Super-populations can be expanded to visualize the next hierarchical level. Left panel: the three super-populations in a collapsed view. Right panel: The three super-populations have been expanded to show their direct children. (C) The panels show the three hierarchical levels of the scene simultaneously. Left: Neuron super-populations at the highest level of abstraction. Middle: The hierarchical entities tree displayed at depth level 2. Right: All entities have been drilled down to show the lowest level of abstraction. Icons have been arranged in a circular layout for convenience for connectivity creation.

Figure 4

Connections and inputs. (A) Connections are created by dragging with the mouse from the source to the target population. The panel on the right shows the parameterizable features. Auto-connections can be created through the context menu that appears when right-clicking on a population. (B) Connectivity simultaneously displayed at three levels of abstraction. Note the connections of superpopulations represent the aggregation of the connections of their descendant populations. (C) Inputs can be created as entities that are external to the hierarchy. Note that inputs appear at every level of abstraction.

Figure 5

Flow of data from visual representation to simulation. The user creates a model in ConGen and exports it as a NeuroML file. The translator parses the NeuroML file and converts the connectivity into the Connection Set Algebra. The populations and inputs are built directly in the simulator. Using the Connection Generation Interface (CGI), connections are generated from the connection generation library and passed to the simulator, which can then start the simulation. A future extension of this workflow will allow simulation results to be processed and passed back to ConGen (dotted line).

Figure 6

Views at different levels of cortical microcircuits. Panel (A) shows microcircuit super-population and the Thalamic super-population. Panel (B) presents the microcircuit super-population where all 8 populations can be seen with their respective connections.

Figure 7

Final view of the complete model considering all populations, connections, and input devices.

Figure 8

Modeling of the co-simulation use case starting with the super-population for the brain region represented by NEST, the super-population for the brain regions represented by TVB and the translator modules which have the task of translating spikes to rates and vice versa.

Figure 9

Views of the multiscale model. (A) Establishing connectivity within the brain regions in the TVB super-population using the Atlas based connectivity. A single connection from Brain_region_0 to Brain_region_67 is used to specify the atlas based connectivity. The connection from region 27 to the rate to spike translator device is also visible at this step. (B) The final whole connectivity setup visualized on the right and the inside view of the cortical microcircuit model on the left.