[Paper Review] Common and Individual Structure of Brain Networks

Lu Wang, Zhengwu Zhang+1 more|arXiv (Cornell University)|Jul 20, 2017

Complex Network Analysis Techniques28 references4 citations

TL;DR

This paper proposes the Multiple GRAph Factorization (M-GRAF) model to identify common and individual-specific brain network structures from replicated binary networks using penalized logistic regression and hierarchical eigenvalue decomposition. The method effectively separates shared connectivity patterns from low-dimensional individual deviations, demonstrating strong performance on Human Connectome Project data with improved interpretability and accuracy in modeling brain connectivity heterogeneity.

ABSTRACT

This article focuses on the problem of studying shared- and individual-specific structure in replicated networks or graph-valued data. In particular, the observed data consist of $n$ graphs, $G_i, i=1,\ldots,n$, with each graph consisting of a collection of edges between $V$ nodes. In brain connectomics, the graph for an individual corresponds to a set of interconnections among brain regions. Such data can be organized as a $V imes V$ binary adjacency matrix $A_i$ for each $i$, with ones indicating an edge between a pair of nodes and zeros indicating no edge. When nodes have a shared meaning across replicates $i=1,\ldots,n$, it becomes of substantial interest to study similarities and differences in the adjacency matrices. To address this problem, we propose a method to estimate a common structure and low-dimensional individual-specific deviations from replicated networks. The proposed Multiple GRAph Factorization (M-GRAF) model relies on a logistic regression mapping combined with a hierarchical eigenvalue decomposition. We develop an efficient algorithm for estimation and study basic properties of our approach. Simulation studies show excellent operating characteristics and we apply the method to human brain connectomics data.

Motivation & Objective

• To address the challenge of identifying shared and individual-specific structures in replicated binary networks, particularly in brain connectomics.
• To develop a statistical model that separates common network topology from low-dimensional individual-specific deviations in brain connectivity.
• To provide a scalable, computationally efficient method for analyzing large-scale brain network data from multiple subjects.
• To enable better prediction of cognitive and behavioral traits through individualized network structure modeling.

Proposed method

• The M-GRAF model uses a logistic regression framework to model edge probabilities in binary networks, incorporating both common and individual-specific latent structures.
• It applies hierarchical eigenvalue decomposition to estimate a shared low-dimensional embedding matrix Q that captures common network structure.
• Individual-specific deviations are modeled via subject-specific latent vectors λik, which are estimated through penalized likelihood with sparsity-inducing constraints.
• An alternating optimization algorithm is used to sequentially estimate the shared basis Q and individual latent vectors, leveraging Rayleigh-Ritz theorem for efficient computation.
• The method includes a spectral decomposition of mean-centered adjacency matrices and applies logistic regression with positivity constraints to ensure valid probability estimates.
• The algorithm is designed for scalability and is validated through simulation studies and real-world application to Human Connectome Project data.

Experimental results

Research questions

• RQ1 How can we effectively separate common and individual-specific structural patterns in multiple replicated binary brain networks?
• RQ2 What statistical model can jointly estimate shared network topology and low-dimensional individual deviations while maintaining interpretability?
• RQ3 How does the proposed M-GRAF model compare to existing methods in terms of accuracy and computational efficiency for large-scale connectomics data?
• RQ4 Can individual-specific network deviations predict cognitive or behavioral traits in human brain networks?
• RQ5 What is the impact of regularization and low-rank structure on the estimation of edge probabilities in binary network data?

Key findings

• The M-GRAF model demonstrates excellent operating characteristics in simulation studies, accurately recovering both common and individual-specific network structures.
• The method successfully identifies shared connectivity patterns across subjects, such as cross-hemisphere connections, which are associated with cognitive traits like visuospatial processing.
• Individual-specific deviations in brain networks, particularly in inter-hemispheric connectivity, show predictive potential for cognitive and behavioral variation.
• The algorithm achieves high computational efficiency through alternating optimization and spectral decomposition, enabling application to large-scale brain network data.
• Empirical results on Human Connectome Project data confirm that the model captures biologically meaningful network heterogeneity while maintaining statistical robustness.
• The use of penalized logistic regression with eigenvalue decomposition ensures valid probability estimates and improves model stability compared to standard spectral methods.