Activity Number:
|
56
- Bayesian Analysis of Functional and Structured Data
|
Type:
|
Contributed
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 2:00 PM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract #313704
|
|
Title:
|
Bayesian Graphical Regression for Multiple Data Types
|
Author(s):
|
Zhenwei Zhou* and Ching-Ti Liu
|
Companies:
|
Boston University and Boston University
|
Keywords:
|
Varying Graph Structure;
Directed Acyclic Graph;
Multiple Data Types;
Extended Rank Likelihood;
Latent Variable Model;
Bayesian Inference
|
Abstract:
|
Integrative analysis of multi-omics data involves the analysis of multi-platform data with various types, such as continuous, binary and ordinal. Modeling the conditional independence structures among different types of variables is often of interest; furthermore, making the structures variable to additional subject-level covariates can be useful in the scenario where the independence structures vary among populations. Bayesian graphical regression has been developed to model the conditional independence structures in the presence of additional subject-level covariates while the usually unmet normality assumption in data with multiple types is required. We generalize this framework utilizing the copula model with extended rank likelihood function that incorporates multiple data types for the estimation of directed graph structure varying flexibly with additional covariates. The structures and covariates can be modeled both linearly and nonlinearly. The sparsity of edges in the graph and the selection of additional covariates are imposed by nonlocal priors. We demonstrate the proposed method by simulation studies and a disease-biomarker relationship study in the Framingham sample.
|
Authors who are presenting talks have a * after their name.