Visual Connectivity Generation
Summary
There is a gap between experimental descriptions of connectivity in neuronal networks and information which can be used by simulators to generate network connectivity. With this project, we aim to generate a visual language for connectivity and a framework which can serve as a bridge between experimental datasets and simulation models. Empirically derived data and theoretical models expressed in a shared visual representation called NeuroScheme are used to generate connections for simulation or further processing.
This solution is simulator independent and open source.
Background and motivation
The study of connectivity is central in the diverse disciplines of neuroscience. Connectivity information can be derived at multiple scales from experimental data and various theoretical models. Structured definitions of network connectivity are also required to construct models for simulation. However, the connectivity information in these two contexts is represented differently. This results in a language gap limiting the flow of knowledge learned at different levels of abstraction. In this work, we present a first step in the creation of a shared visual language to bridge this gap between model based and empirical neuroscience, allowing us to work towards a single integrated representation of the brain.
Project description
We have developed a visual and source-agnostic interactive interface to generate connectivity in neural networks at various scales. Based on NeuroScheme [1] and the Connection Set Algebra (CSA)[2], we can generate connectivity and use it in simulator-specific scripts to later perform simulations of the dynamics of the network. Our approach allows us to interactively create, explore and visualize connectivity even for large scale networks where probabilistic connections are used to describe synapse generation.
With this approach, we offer the neuroscientific community a generic tool for easy generation and exploration of connectivity. The lack of dependency on a specific simulator makes this tool a good starting point for validation of complex neural network models using many simulation and emulation platforms, particularly when coupled. Our future applications involve incorporating this tool to complete workflows consisting of raw data processing, interactive exploration, creation and visualization of abstract connectivity models, simulation, analysis and validation.
Figure 1: Visualization of a cortical microcircuit model using NeuroScheme and the ViCoGen connectivity framework. Extracted from [3].
Our contribution
The SimLab Neuroscience works on the definition of use cases, setups and scientific requirements. The implementation of the translation from visual language to simulation scripts is the subject of the Master’s thesis “Generating Neural Network Connectivity from a Visual Representation” by Patrick Herbers, written in 2017 at Ruhr-Universität Bochum under the guidance of the SimLab Neuroscience.
Our collaboration partners
In this project we are collaborating with RWTH Aachen University, the Universidad Rey Juan Carlos and the Center for Computational Simulation of the Universidad Politécnica de Madrid in Spain.
References
- Pastor, Luis, et al. "Neuroscheme: efficient multiscale representations for the visual exploration of morphological data in the human brain neocortex." (2015).
- Djurfeldt, Mikael. "The connection-set algebra—a novel formalism for the representation of connectivity structure in neuronal network models." Neuroinformatics 10.3 (2012): 287-304.
- Herbers, P. (2017, November). Generating Neural Network Connectivity from a Visual Representation. Master’s thesis, Applied Informatics, Univ. of Bochum, Germany. https://www.ini.rub.de/upload/file/1527838547_f919457ea2789d6cbc5d/masters_thesis_herbers_2017.pdf
