Combining network topology and information theory to construct representative brain networks.

Area of Science:

• Neuroscience
• Network Science
• Graph Theory

Background:

• Network neuroscience uses graph theory to study the brain as a complex network.
• Inconsistent network construction pipelines hinder the generalizability of findings in brain network research.
• A wide variety of pipelines in the literature make cross-study comparisons challenging.

Purpose of the Study:

• To develop a method for generating brain networks that are maximally representative of networks derived from the same neuroimaging data.
• To address the issue of non-generalizable graph-theoretical results due to idiosyncratic network construction pipelines.
• To identify optimal node definition and thresholding procedures for deriving representative brain networks.

Main Methods:

• Employed portrait divergence, an information-theoretic measure, to minimize divergence between network topologies.
• Utilized functional and diffusion MRI data from the Human Connectome Project.
• Investigated anatomical, functional, and multimodal parcellations at multiple scales with 48 edge definition methods.

Main Results:

• Achieved highest network representativeness using parcellations of approximately 200 regions.
• Identified efficiency-cost optimization for filtering functional networks as a key factor.
• Highlighted suitable alternative methods for network construction.

Conclusions:

• Specific node definition and thresholding procedures can yield representative brain networks from human neuroimaging data.
• Minimizing portrait divergence is a viable strategy for enhancing the generalizability of network neuroscience findings.
• The findings provide practical guidelines for neuroscientists to derive robust and comparable brain network models.