Title: Variable selection in mixture models: Uncovering cluster structures and relevant features
Authors: Mahlet Tadesse - Georgetown University (United States) [presenting]
Abstract: Identifying latent classes and component-specific relevant predictors can shed important insights when analyzing high-dimensional data. Methods to address this problem in a unified manner will be presented by combining mixture models and variable selection ideas in different contexts. In particular, we will discuss (1) a bi-clustering approach that allows clustering on subsets of variables by introducing latent variable selection indicators in finite or infinite mixture models, (2) an integrative model to relate two high-dimensional datasets by fitting a multivariate mixture of regression models using stochastic partitioning, and (3) a mixture of regression trees approach to uncover homogeneous subgroups and their associated predictors accounting for non-linear relationships and interaction effects. We will illustrate the methods with various genomic applications.