Automated customization of large-scale spiking network models to neuronal population activity

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neural Engineering

Background:

• Computational models are crucial for understanding brain function by simulating neural activity.
• Spiking neural networks (SNNs) model neuronal biophysics but are complex to parameterize.
• Current heuristic methods limit SNNs' ability to match large-scale neural recordings.

Purpose of the Study:

• To introduce an automated method, Spiking Network Optimization using Population Statistics (SNOPS), for configuring SNNs.
• To enable SNNs to reproduce population-level neural activity statistics from recordings.
• To enhance the discovery of complex neural activity regimes.

Main Methods:

• Developed the SNOPS algorithm for automatic SNN parameter optimization.
• Validated SNOPS by confirming accurate recovery of simulated neural activity statistics.
• Applied SNOPS to large-scale neural recordings from macaque visual and prefrontal cortices.

Main Results:

• SNOPS successfully reproduced population-wide covariability from neural recordings.
• The application of SNOPS revealed previously unrecognized limitations in current SNN models.
• The method demonstrated high accuracy in recovering simulated neural activity statistics.

Conclusions:

• SNOPS provides an automated and effective approach for customizing SNNs.
• This method facilitates the development of more accurate and biologically plausible neural network models.
• SNOPS offers a pathway to deeper insights into neural computation and brain function.