B0734
Title: Heterogeneous block covariance model for community detection
Authors: Xiang Li - The George Washington University (United States)
Qing Pan - George Washington University (United States)
Ning Hao - University of Arizona (United States)
Yunpeng Zhao - Colorado State University (United States) [presenting]
Abstract: Community detection is a clustering method based on the pairwise relationships of objects, such that objects classified in the same group are more densely connected than objects from different groups. While most model-based community detection methods, such as the stochastic block model and its variants, are designed for networks with binary (yes/no) edges, many practical scenarios involve edges with continuous weights that reflect different degrees of connectivity. The heterogeneous block covariance model (HBCM) introduces a novel clustering structure on the covariance matrix, where edges possess signed and continuous weights. The HBCM considers the heterogeneity of objects when forming connections within a community. It proposes a novel variational expectation-maximization (EM) algorithm to estimate the group membership. The HBCM provides provably consistent estimations of clustering memberships, and its superior performance is observed in numerical simulations with various setups. The model is then applied to a yeast gene expression dataset to detect gene clusters regulated by different transcript factors during the yeast cell cycle.