Bayesian LASSO for Non-Stationary Gaussian Linear Mixed Effects Model

• Activity Number: 347 - Computationally Intensive Bayesian Methodology
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Emrah Gecili* and Siva Sivaganesan and Assem G Ziady and Rhonda Szczesniak
• Companies: Cincinnati Children's Hospital Medical Center and University of Cincinnati and Cincinnati Children's Hospital Medical Center and Cincinnati Children's Hospital
• Keywords: Bayesian lasso; proteomic data; linear mixed effects model

Abstract

In proteomics experiments, variable selection can involve thousands of protein isoforms observed from a relatively small number of samples. Frequentist and Bayesian Lassos have been suggested for variable selection in omics data, but most approaches consider linear regression modeling and ignore correlation arising from repeated measurements. Our work is motivated by the need to identify a small set of protein isoforms that improve the prediction of rapid lung-function decline for individuals with cystic fibrosis lung disease. We propose a Bayesian Lasso for a Gaussian linear mixed effects model with non-stationary covariance to account for the complicated structure of longitudinal lung-function data while simultaneously estimating unknown parameters and selecting important protein isoforms to improve our prediction model. Different choices of shrinkage priors that have been used by others are evaluated to induce variable selection in a fully Bayesian framework. We apply the proposed method to real proteomic and lung-function data. Methodology and results are discussed.