|
Activity Number:
|
127
- Statistical Methods for Multi-Omic Data Analysis
|
|
Type:
|
Topic-Contributed
|
|
Date/Time:
|
Monday, August 9, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Section on Statistics in Genomics and Genetics
|
|
Abstract #317392
|
|
|
Title:
|
Bayesian Inference on Multilayered Non-Normal Graphical Models
|
|
Author(s):
|
Min Jin Ha* and Moumita Chakraborty and Anindya Bhadra and Veerabhadran Baladandayuthapani
|
|
Companies:
|
University of Texas MD Anderson Cancer Center and University of Texas MD Anderson Cancer Center and Purdue University and University of Michigan
|
|
Keywords:
|
multilayered Gaussian graphical models;
multiomics;
Bayesian;
non-normal
|
|
Abstract:
|
Integrative analysis of data generated from multiomic platforms for modeling dependences across various domains is an important step toward the development of successful therapeutic strategies in cancer. Multilayered Gaussian graphical models (mlGGMs) have been proposed to characterize conditional dependence structures of such multi-level continuous data, where the variables are naturally partitioned into multiple ordered layers. The mlGGMs in multiomics characterize both directed and undirected edges for inter- and intra- platform dependencies, respectively. However, Gaussian assumption is often inappropriate due to non-normal marginal distributions coming from heavy tails, potentially leading to inaccurate graphical structures. We propose a robust Bayesian node-wise selection (robust-BANS) approach based on transformations of data using Gaussian scale mixtures to account for the non-normality of each node in continuous multivariate data. This flexible Bayesian modeling strategy facilitates identification of conditional sign dependences among nodes that are not normal while still being able to infer conditional dependences among nodes that are normal.
|
Authors who are presenting talks have a * after their name.
