|
Activity Number:
|
98
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 12, 2002 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences*
|
| Abstract - #300601 |
|
Title:
|
Association Measures and Tree Models for Phenotyping Using Gene Expression Profiles
|
|
Author(s):
|
Mike West*+ and Jennifer Pittman
|
|
Affiliation(s):
|
Duke University and Duke University
|
|
Address:
|
Duke University, Durham, North Carolina, 27708-0251, USA
|
|
Keywords:
|
Gene expression ; Molecular phenotyping ; Predictive discrimination ; Bayesian tree models ; Association analysis
|
|
Abstract:
|
The exploration and modelling of relationships between expression patterns of many, many genes and clinical outcomes or physiological states presents major challenges to statistical modelling. This talk discusses some of these challenges, including: the development of predictive models; issues of variable selection and the often acute sensitivity of model predictions to small changes in data configurations and variables selected; the need for flexible, non-linear measures of association between gene expression profiles and outcomes; and the integration of gene expression data with other covariates in predictive models. The methodological heart of the talk revolves around some of our recent work on novel, non-linear association measures for binary outcomes, and associated novel approaches to predictive phenotyping that utilize new methods for constructing Bayesian classification trees. Some examples in analysis of gene expression profiles will be drawn from current collaborations in breast cancer and cardiovascular disease studies.
|
