Multi-subject Stochastic Blockmodels for adaptive analysis of individual differences in human brain network cluster...

Area of Science:

• Neuroscience
• Network Science
• Statistical Modeling

Background:

• Elucidating brain network cluster structure is crucial but challenging for multi-subject data.
• Existing methods struggle with subject-by-subject analysis or group-averaged networks, limiting group comparisons and variability assessment.

Purpose of the Study:

• To propose novel extensions of the Stochastic Blockmodel (SBM) for multi-subject brain network analysis.
• To develop Multi-Subject Stochastic Blockmodels (MS-SBMs) that account for between-subject variability and covariate effects.

Main Methods:

• Introduced two MS-SBMs using mixture and regression models to estimate cluster structures and covariate effects.
• Evaluated MS-SBMs using synthetic data and compared them against Fast Louvain and Newman Spectral algorithms.
• Applied MS-SBMs to resting-state fMRI data from healthy volunteers and individuals with schizophrenia.

Main Results:

• MS-SBMs accurately recovered synthetic network cluster structures, outperforming standard modular decomposition methods.
• Permutation tests based on MS-SBM parameters demonstrated robust statistical inference and Type I error control.
• The Heterogeneous Stochastic Blockmodel (Het-SBM) identified diverse network topologies, including modular and core structures, in fMRI data.

Conclusions:

• MS-SBMs provide a flexible and accurate framework for analyzing multi-subject brain network structures.
• These models effectively capture individual differences and improve statistical inference in network neuroscience.
• The proposed methods offer advancements for understanding brain network variability in health and disease.