053
Bayesian prediction combining genotyped and non-genotyped individuals

Monday, August 18, 2014: 10:30 AM
Stanley Park Ballroom (The Westin Bayshore)
Dorian J. Garrick , Iowa State University, Ames, IA
Jack C. M. Dekkers , Iowa State University, Ames, IA
Bruce L Golden , Calpoly, San Luis Obispo, CA
Rohan L Fernando , Iowa State University, Ames, IA
Abstract Text: Conventional pedigree- and performance-based national evaluations typically involve hundreds of thousands if not millions of animals.  But only a small proportion of individuals with performance records have typically been genotyped to date.  Bayesian methods have been widely adopted for analysis of these genotyped individuals, but implementation typically involves two-step approaches to blend genomic predictions on genotyped individuals with information from conventional analyses for non genotyped animals. Here we present a Bayesian approach that extends commonly-used methods including BayesA, BayesB, BayesC, and BayesCπ, to a single step method using observations from all genotyped and non genotyped individuals.  Unlike single-step GBLUP, our approach does not require direct inversion of any matrices and is well suited to parallel computing approaches.

Keywords:

Genomic prediction