Integrative learning for population of dynamic networks with covariates

• Published in: NeuroImage (Orlando, Fla.) Vol. 236; p. 118181
• Main Authors: Kundu, Suprateek, Ming, Jin, Nocera, Joe, McGregor, Keith M.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.08.2021; Elsevier Limited; Elsevier
• Subjects: Aerobics; Aged; Aged, 80 and over; Algorithms; Bayesian analysis; Brain - diagnostic imaging; Brain - physiology; Brain mapping; Computer Simulation; Connectome; Dynamic networks; EM Algorithm; Estimates; Exercise; Female; Functional magnetic resonance imaging; Humans; Integrative learning; Magnetic Resonance Imaging; Male; Middle Aged; Mixture models; Nerve Net - diagnostic imaging; Nerve Net - physiology; Neural networks; Neural Networks, Computer; Performance evaluation; Physical fitness; Population; Unsupervised Machine Learning; Dynamic networks; EM Algorithm; Integrative learning; Mixture models
• ISSN: 1053-8119; 1095-9572; 1095-9572
• DOI: 10.1016/j.neuroimage.2021.118181

•One of the first integrative methods in literature to jointly estimate a collection of unknown dynamic networks for a group of individuals, guided by covariate information, which involves significant methodological and computational novelty. We develop the approach for both dynamic pairwise correlations and partial correlations.•The proposed method is able to systematically borrow information across heterogeneous samples in an unsupervised manner in order to compute individual-specific dynamic networks. This results in considerable gains in estimation accuracy by leveraging common connectivity patterns•In addition to estimating individual level networks, the method is able to identify subgroups of individuals with similar dynamic connectivity patterns and report subgroup level network characteristics that are more robust to noise and heterogeneity in the samples.•Extensive numerical studies illustrate significant gains due to incorporation of covariate information, and also relative to existing single-subject and multi-subject dynamic connectivity approaches, in terms of recovering the true dynamic network structure.•Our analysis of fMRI block task data illustrated the considerable advantages of systematically borrowing information across samples under the proposed approach compared to existing dynamic connectivity methods. Alternate single-subject analyses fail to detect connectivity change points that is biologically impractical in a block task experiment. Although there is a rapidly growing literature on dynamic connectivity methods, the primary focus has been on separate network estimation for each individual, which fails to leverage common patterns of information. We propose novel graph-theoretic approaches for estimating a population of dynamic networks that are able to borrow information across multiple heterogeneous samples in an unsupervised manner and guided by covariate information. Specifically, we develop a Bayesian product mixture model that imposes independent mixture priors at each time scan and uses covariates to model the mixture weights, which results in time-varying clusters of samples designed to pool information. The computation is carried out using an efficient Expectation-Maximization algorithm. Extensive simulation studies illustrate sharp gains in recovering the true dynamic network over existing dynamic connectivity methods. An analysis of fMRI block task data with behavioral interventions reveal sub-groups of individuals having similar dynamic connectivity, and identifies intervention-related dynamic network changes that are concentrated in biologically interpretable brain regions. In contrast, existing dynamic connectivity approaches are able to detect minimal or no changes in connectivity over time, which seems biologically unrealistic and highlights the challenges resulting from the inability to systematically borrow information across samples.