Improved functional connectivity network estimation for brain networks using multivariate partial coherence.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Graphical networks and metrics are vital for characterizing brain networks and function using electrophysiological data.
• Current methods often rely on pair-wise correlation, which may not fully capture complex neural interactions.

Purpose of the Study:

• To demonstrate the superior accuracy of partial correlation over pair-wise correlation for estimating functional brain networks.
• To compare network metrics derived from conditional (partial coherence) and unconditional (pair-wise correlation) graphical networks.

Main Methods:

• Constructed graphical networks using coherence (unconditional) and multivariate partial coherence (MVPC) (conditional) estimates.
• Calculated binary and weighted network metrics including node degree, path length, clustering coefficients, and small-world index.
• Applied methods to simulated cortical neuron data and experimental single-unit recordings from a rat epilepsy model.

Main Results:

• Conditional network metrics provided a more accurate representation of known connectivity in simulated data.
• Conditional node degree (2-13) was more precise than unconditional node degree (6-80).
• Experimental data showed similar trends: lower binary node degree and longer binary path lengths with conditional networks.

Conclusions:

• Conditional networks, by accounting for common dependencies via partial coherence, offer a more accurate graphical network model of neural interactions.
• Results advocate for using partial correlation (MVPC) instead of pair-wise correlation for functional brain network analysis.