多分类计数数据的回归建模及其贝叶斯推断

Regression Modelling and Bayesian Inference for Multivariate Categorical Count Data

本文使用多项-Logistic-正态模型分析多分类计数响应数据,基于贝叶斯推断方法研究模型的估计及其变量选择方法。通过引入Polya-Gamma分布的潜在变量和回归系数的“Spike-and-Slab”先验,得到了Gibbs后验抽样算法。数值模拟研究和RNA-seq基因实例数据分析验证了MLN模型及其贝叶斯推断方法的有用性。

Inthis paper,Multinomial-Logistic-Normal model is proposed to analyze multivariate categorical count response data,and the estimation and variable selection method are investigated based on Bayesian inference approach.We obtain the Gibbs posterior sampling algorithm by introducing the latent variable with Polya-Gamma distribution and Spike-and-Slab prior for regression coefficients.Simulation studies and RNA-seq gene real data analysis further verify the usefulness of MLN model and the proposed Bayesian inference procedure.