|
Activity Number:
|
198
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #301256 |
|
Title:
|
The Geno-Pheno Problem: Predicting Fitness from Genotype Using the Exchangeable within Subsets Prior and Prior Model Selection
|
|
Author(s):
|
Christina R. Kitchen*+ and Robert E. Weiss and Gang Liu
|
|
Companies:
|
University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles
|
|
Address:
|
650 Charles E. Young Dr., Los Angeles, CA, 90095-1772,
|
|
Keywords:
|
HIV ; viral fitness ; hierarchical modeling ; Bayes theorem
|
|
Abstract:
|
The genotype-phenotype problem is the fundamental problem of genetics where one utilizes the genotype (genetic information) to predict phenotype (observable characteristic). Our phenotype is viral fitness defined as the capacity of the virus to replicate as compared to a standard. We construct models to predict viral fitness as a function of mutations from a standard wild-type virus. Data of this nature are difficult to analyze because there are potentially many more parameters than observations. We treat the problem as a regression problem with both a shrinkage component and a variable selection component. Using the scientific literature, we set up several informative priors that treat parameter subsets as exchangeable within subsets (the ESP prior). There are several scientific papers, each providing separate prior information. We then do prior model selection (PMS) to select the best prior. In this way we provide a solution for the genotype-phenotype problem. Viral fitness and genotypic data from 161 clade B HIV-1 viral sequences were obtained to predict the level of viral fitness as a function of mutation.
|
