Kernel machine tests of association using extrinsic and intrinsic cluster evaluation metrics

Area of Science:

• Neuroscience
• Network Science
• Biostatistics

Background:

• Mesoscale human brain network topology varies with cognition, disease, and age.
• Existing algorithms detect brain communities but lack robust inference methods.
• Current analysis compares community detection within subjects, limiting generalizability and covariate inclusion.

Purpose of the Study:

• To develop a novel semiparametric kernel machine regression model for brain network community analysis.
• To enable inference on the association between brain community structures and outcomes, accommodating covariates.
• To generalize brain network analysis to non-linear spaces using similarity-based kernels.

Main Methods:

• Proposed a semiparametric kernel machine regression model for continuous or binary outcomes.
• Modeled covariate effects parametrically and brain connectivity nonparametrically.
• Incorporated similarity measures between network community structures into a kernel distance function.

Main Results:

• The methodology was evaluated on both simulated and real brain network datasets.
• Demonstrated the ability to generalize high-dimensional brain network features to non-linear spaces.
• Facilitated a wider class of distance-based algorithms for network analysis.

Conclusions:

• The proposed model offers a flexible framework for inferring associations between brain network communities and outcomes.
• This approach addresses limitations in current methods by accommodating covariates and enabling broader statistical analysis.
• The study advances the analysis of brain network structures in neuroscience and related fields.