|
Activity Number:
|
19
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308765 |
|
Title:
|
Latent Class Modeling of Growth and Reproduction in Sublethal Toxicity Tests
|
|
Author(s):
|
Michael Pennell*+ and David B. Dunson
|
|
Companies:
|
The Ohio State University and National Institute of Environmental Health Sciences
|
|
Address:
|
320 West 10th Avenue, Columbus, OH, 43210,
|
|
Keywords:
|
Bayesian methodology ; Dirichlet process ; Latent variables ; Mixture model ; Order restrictions ; Toxicology
|
|
Abstract:
|
Medium-throughput methods have been established to rapidly screen many developmental toxins using Caenorhabditis elegans. In these studies, organisms are exposed to one of 12 doses and their size is measured after 72 hours, during which time unaffected organisms may also reproduce. This presents some interesting statistical challenges as 1.) there is no way of distinguishing the progeny from the original experimental units and 2.) there is no consensus on how to define larval stage in terms of size. To address these issues, we propose a latent class model for growth and reproduction. We model the number of organisms at each dose using a Dirichlet process mixture of Poissons and size using a mixture of two multivariate t distributions. We also use mixture priors to test for changes across dose with respect to growth and reproduction and apply order restrictions to increase efficiency.
|
