A Latent Factor Model with a Mixture of Sparse and Dense Components for Gene Expression Data with Confounders

• Activity Number: 694
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Chuan Gao*+ and Barbara Engelhardt
• Companies: Institute for Genome Sciences and Policy and Duke University
• Keywords: sparse factor analysis ; latent dirichlet allocation ; stick breaking ; dictionary learning ; gene expression ; mixture models

Abstract

Sparse latent factor models are useful in terms of extracting interpretable features and feature relationships from high dimensional data. Interpretability is especially important for problems in genome sciences, where high-dimensional sequencing data are generated at an exploding rate. One important problem in genome science is to understand the relationship between genes from gene expression data, where there may be substantial technical noise. To address this problem, we developed a Bayesian sparse latent factor model that uses a three parameter beta prior to flexibly model shrinkage on the loading matrix. We further use a two-component mixture to model each factor loading as being generated from a sparse or dense mixture component; this allows some dense factors that represent confounding factors and are not expected to be sparse and some sparse factors representing local gene interactions. We tested our model on simulated data and found that we successfully recover the true latent structure. We applied the model to gene expression data and found that it identifies known covariates and small groups of co-regulated genes that can then be used for downstream analyses.