Inferring structural connectivity using Ising couplings in models of neuronal networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Functional connectivity metrics are crucial for inferring neuronal structural connectivity.
• Maximum entropy Ising models show promise in identifying direct neuronal interactions by discounting indirect effects.
• A lack of benchmarking exists for Ising model performance against other metrics at the microscopic neuronal network scale.

Purpose of the Study:

• To benchmark Ising model couplings against partial and cross-correlations for inferring synaptic connectivity.
• To evaluate performance across diverse in silico neuronal network conditions, including correlation levels, firing rates, sizes, densities, and topologies.

Main Methods:

• In silico simulation of neuronal networks.
• Comparison of Ising model couplings, partial correlations, and cross-correlations.
• Systematic variation of network parameters: correlation levels, firing rates, network size, density, and topology.

Main Results:

• The performance hierarchy of functional connectivity metrics is primarily dictated by network correlation levels.
• Ising couplings excel at detecting structural links in networks with very weak correlations.
• Partial correlations demonstrate superior performance over Ising couplings and cross-correlations in highly correlated networks.
• Findings remain consistent across variations in firing rates, network sizes, and topologies.

Conclusions:

• The choice of functional connectivity metric significantly impacts the accuracy of structural connectivity inference.
• Ising models are most effective for neuronal networks with low correlation levels.
• Partial correlations are recommended for networks exhibiting strong correlations.
• This study provides essential guidance for selecting appropriate tools for reconstructing neuronal structural connectivity.