Flexible Bivariate Correlated Count Data Regression with Application in Gene Coexpression Analysis Based on RNA-Sequencing Data

• Activity Number: 415 - Statistical Methods for Gene Expression and RNA-Seq Analysis
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Zichen Ma*
• Companies: University of South Carolina
• Keywords: Bivariate count data; Bivariate regression; RNA-sequencing data; Bayesian random effects model

Abstract

There has been a growing interest in modeling multivariate correlated count data due to its importance in applications such as gene co-expression based on RNA-sequencing data. Specifically, one may aim to quantify “liquid association”, that is, the change in the correlation between responses given various values of some covariates. We propose three distinct Bayesian approaches to modeling bivariate correlated count data, aiming to simultaneously regress the mean, dispersion, and the correlation onto covariates. The first approach utilizes an explicit bivariate negative binomial likelihood provided in Famoye (2010). The second approach is a bivariate generalization of the univariate Poisson-gamma mixture model. And the third approach is a Gaussian copula model. We use simulation to compare the three approaches in terms of model fitting based on the log-pseudo marginal likelihood, and identifying liquid association based on the power of associated hypotheses. We illustrate the proposed approaches using breast cancer data from the International Cancer Genome Consortium data portal.