|
Activity Number:
|
421
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309438 |
|
Title:
|
A Bayesian Model for Genetic Association with Phenotype Defined in a Limited Range
|
|
Author(s):
|
Ling Wang*+ and Vikki Nolan and Clinton Baldwin and Martin Steinberg and Paola Sebastiani
|
|
Companies:
|
Boston University and Boston University and Boston University and Boston University and Boston University
|
|
Address:
|
401 Broadway, Cambridge, MA, 02139,
|
|
Keywords:
|
Bayesian ; DIC ; Beta distribution ; SNP
|
|
Abstract:
|
The phenotype from genetic association studies is often a continuous variable such as a probability, or a percentage, and are defined in [0,1]. Linear regression with the transformed phenotype as dependent variable may lead to false positive associations and produce poor fit due to the transformation. We propose a Bayesian approach that uses Beta distribution to model the phenotype in the correct range of definition. We estimate the association with or without adjustment for confounding in a Bayesian framework using MCMC computation, and evaluate this procedure by assessing the false positive rates (FPR) and true positive rates (TPR) in simulated data. Compared with the model assuming a lognormal distribution, our method has a comparable FPR but a higher TPR. We apply this method to candidate-gene analysis that might impact fetal hemoglobin concentration in sickle cell anemia patients.
|
