Leveraging edge-centric networks complements existing network-level inference for functional connectomes

Area of Science:

• Neuroscience
• Network Science
• Computational Biology

Background:

• The human connectome exhibits modular organization into large-scale brain networks.
• Network inference methods traditionally rely on node-centric network construction.
• Novel edge-centric networks, based on edge similarity, offer an alternative construction approach.

Purpose of the Study:

• To evaluate the efficacy of edge-centric networks in network-level inference.
• To compare edge-centric network inference with traditional node-centric methods and edge-level statistical approaches.
• To assess the power, specificity, and consistency of edge-centric networks across diverse datasets and sample sizes.

Main Methods:

• Utilized data from the Human Connectome Project, Healthy Brain Network, and Philadelphia Neurodevelopmental Cohort.
• Employed a resampling technique with varying sample sizes (n=40, 80, 120).
• Compared network-level inference using edge-centric networks against node-centric networks, edge-level FDR correction, and NBS.

Main Results:

• Edge-centric networks demonstrated superior performance in inference, edge-level FDR correction, and NBS compared to node-centric networks across all datasets and sample sizes.
• Edge-centric networks showed greater consistency in clustering similar effect sizes.
• Node-centric networks occasionally exhibited lower average within-network effect size variability.

Conclusions:

• Edge-centric networks provide a powerful and accurate method for network-level inference in connectomics.
• This approach complements existing inferential techniques by offering robust analysis of brain networks.
• The findings support the utility of edge-centric network construction for advancing neuroimaging analysis.