Genetically Informed Brain Parcellation Through Structured Multi-Task Modeling

• 成果类型: Article; Early Access
• 署名作者: Yao, Yisha; Hu, Yue; Wang, Shiying; Dai, Wei; Liu, Zihuan; Zhang, Heping
• 署名单位: Columbia University; Yale University; Yale University
• 刊物名称: JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
• ISSN/ISSBN: 0162-1459; 1537-274X
• DOI: 10.1080/01621459.2026.2657607
• 发表日期: 2026-06-20
• 关键词: brain networks; em algorithm; genetics; minimax concave penalty; Shrinkage prior; variable selection; group lasso; regression; mixtures

摘要

The organization of human brain subnetworks is fundamental to understanding cognition and neuropsychiatric health. Existing approaches predominantly construct subnetworks by clustering brain regions according to measured imaging phenotypes or functional correlation. Although successful, such phenotype-based parcellations reflect composite effects of genetics, environment, lifestyle, and measurement noise, thereby limiting biological interpretability and obscuring subnetworks attributable to specific mechanisms. Accumulating evidence suggests that brain connectivity is substantially heritable and genetic influences are regionally heterogeneous and aligned with functional architecture. Motivated by these findings, we propose a novel statistical framework to construct genetically informed brain subnetworks by disentangling genetic effect from other sources of variation. The framework integrates two key components: a high-dimensional multi-task learning model that decomposes composite effects, and a group-wise mixture structure on gene-specific regional effects to identify latent clustering patterns induced by individual genes. This formulation enables direct modeling of regional genetic influences rather than relying on aggregate heritability measures, yielding subnetworks with shared molecular mechanisms. We further develop an iterative algorithm for model fitting, and establish theoretical guarantees for its efficiency, supported by simulation results. This framework is general and can be extended to derive subnetworks induced by other factors of interest. We implement the method on a dataset from the UK Biobank (UKB) for discovery, and validate the results on the Human Connectome Project Young Adult (HCP-YA) dataset. Our results reveal intrinsic, heritable brain atlases that complement conventional phenotype-based parcellations and provides new insight into the interplay among genes, brain organization, and neuropsychiatric disorders. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.