|
Activity Number:
|
609
- New Advances in Analysis of Complex Cohort Studies
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Thursday, August 3, 2017 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Korean International Statistical Society
|
|
Abstract #323933
|
|
|
Title:
|
Nonparametric Conditional Graphical Models with High-Dimensional Predictors
|
|
Author(s):
|
Yi Li*
|
|
Companies:
|
University of Michigan
|
|
Keywords:
|
Nonparametric Conditional Graphical Models ;
non-Gaussian response network ;
reproducing kernel Hilbert space ;
screening ;
high-dimensional data analysis
|
|
Abstract:
|
Understanding how genes interplay with each other and how their regulations are associated with ultrahigh-dimensional genomic markers helps uncover the underlying mechanism of disease progression process. Conditional Gaussian graphical models are commonly used in simultaneously learning the structure of response network and its associations with extraneous predictors. However, most of them can not be directly applied to non-Gaussian response networks or be used to detect associations between response network edges and predictors. These limit their usage in modern biomedical ``-omic" studies. We propose a non-parametric conditional graphical model by embedding it in a reproducing kernel Hilbert space (RKHS). We also propose a graphic screening method for variable selection and network recovery. We show that the proposed method can consistently select important predictors and recover the response network structure. The proposed method is computationally inexpensive and can be directly applied to analyze ``-omic" scaled networks and DNA data. The proposed method is applied to the cancer genome atlas (TCGA) data to study the cancer-triggering biological pathways.
|
Authors who are presenting talks have a * after their name.
