A random covariance model for bi-level graphical modeling with application to resting-state fMRI data

Area of Science:

• Computational Statistics
• Network Science
• Biostatistics

Background:

• Bi-level graphical modeling involves inferring group-level and individual-level network structures simultaneously.
• Applications include multi-subject neuroimaging and genomics, requiring methods that capture both shared and unique network characteristics.
• Existing methods often focus solely on individual-level models, neglecting the underlying group structure.

Purpose of the Study:

• To propose a novel and efficient statistical method, the random covariance model, for bi-level graphical modeling.
• To simultaneously learn group- and individual-level graphical models.
• To provide a measure of degrees-of-freedom for model complexity and selection.

Main Methods:

• Developed a random covariance model, analogous to random effects models for mean structures.
• The method accounts for similarities between individual graphical models.
• Infers shared group-level connections and individual-level networks concurrently.

Main Results:

• The random covariance model efficiently learns both group and individual graphical models.
• Demonstrated asymptotic properties and finite-sample performance via simulations.
• Successfully applied to multi-subject resting-state functional magnetic resonance imaging (fMRI) data from schizophrenia patients.

Conclusions:

• The random covariance model offers a computationally efficient approach for bi-level graphical modeling.
• It effectively identifies both shared group-level functional connectivity and individual-specific networks.
• The method holds significant promise for analyzing complex multi-subject biological and medical data.