Survival Prediction and Variable Selection with Simultaneous Shrinkage and Grouping Priors for Gene Expression Microarray Data

• Activity Number: 637
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: International Indian Statistical Association
• Author(s): Sounak Chakraborty*+ and Kyu Ha Lee and Jianguo Sun
• Companies: University of Missouri-Columbia and Harvard School of Public Health and University of Missouri-Columbia
• Keywords: elastic net ; fused lasso ; grouped lasso ; Bayesian variable selection ; Cox Proportional hazard model ; microarrays

Abstract

In this article, we propose Bayesian penalized regression models for high-dimensional survival data. In the analysis of gene expression data, it is naturally assumed that genes are grouped according to some underlying process. Our proposed framework is motivated by the need of grouped shrinkage estimation to take such consideration into account. Special shrinkage priors correspond to the elastic net, group lasso, and fused lasso penalties are used to incorporate the grouping effect of the gene expression microarray covariates. We adopted Bayesian Cox proportional hazards model where the cumulative baseline hazard function is modeled through a discrete gamma process prior. In the proposed Bayesian approach, the amount of grouped shrinkage are adaptively controlled by estimating tuning parameters via Markov chain Monte Carlo (MCMC) sampling method. The proposed methodologies are very useful when we want to incorporate the cluster structure of gene expression data into the models. We assess the prediction performance of our Bayesian penalized regression methods using simulations and three different real life high-dimensional survival data sets.