Model Selection Strategy for Multiple Binary Trait Loci

• Activity Number: 374
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Cynthia J. Coffman*+ and Krista Nichols and Christine Woods and Wendy Czika and Katy L. Simonsen and Russell D. Wolfinger and Lauren M. McIntyre
• Companies: Center for Health Services Research in Primary Care and Washington State University and SAS Corporation and SAS Corporation and Purdue University and SAS Institute, Inc. and Purdue University
• Keywords: binary trait loci ; model selection ; binary

Abstract

Modern molecular technology coupled with statistical methodology has led to the successful detection and location of QTL as well as single genomic regions (binary trait locus, "BTL") associated with binary traits. For example, single marker associations with the binary trait Ceratomyxa shasta resistance in Oncorhynchus mykiss, rainbow, and steelhead trout, suggest that at least four different genomic loci are associated with the resistance trait. In this work, we present a general probability model developed for BCn and Fn populations for multiple BTL. The probability model leads to a straightforward general likelihood model for an arbitrary number of loci. Model selection criteria (AIC and BIC) are then applied to a set of likelihood models to determine the number of loci associated with a binary trait. This procedure has been evaluated via extensive simulations. A SAS procedure, PROC BTL, implements the multiple loci likelihood models and estimates parameters. We apply our multiple loci binary models for varying numbers of loci to the rainbow trout data using PROC BTL and present our model selection strategy for determining the number of loci associated with the resistance trait.