P0035
BLUP Alphabet for Genomic Prediction

Zhiwu Zhang , Washington State University, Pullman, WA
Jiabo Wang , Northeast Agricultural University, Haerbin, China
Zhengkui Zhou , Chinese Academy of Sciences, Beijing, China
Zhe Zhang , South China Agricultural University, Guangzhou, China
Hui Li , Northeast Agricultural University, HARBIN, China
Di Liu , Heilongjiang Academic of Agricultural science, HARBIN, China
Qin Zhang , China Agricultural University, Beijing, China
Peter Bradbury , Cornell University/USDA-ARS, Ithaca, NY
Rory Todhunter , Cornell University, Ithaca, NY
Edward S. Buckler , USDA-ARS-Cornell University, Ithaca, NY
Prediction of individuals’ genetic merit plays the key role in modern breeding of animals and plants. By using pedigree to define the kinship among individuals, Dr. Charles R. Henderson developed the Best Linear Unbiased Prediction (BLUP) method in 1940-50’s to predict individuals’ genetic effects. The method soon became the widely used standard for animal and plant breeding in 1960-80’s. In early 1990’s, molecular genetic markers were introduced for Marker Assisted Selection (MAS). Later on, a set of markers covering entire genome was introduced by Dr. Rex Bernardo in 1994 to calculate the kinship among individuals as substitution of the pedigree based kinship to derive genomic BLUP. The method is now commonly known as gBLUP. In 2001, the MAS approach was extended by Dr. T. H. E. Meuwissen et al. to substitute few influential markers as fixed effects with all available markers as random effects. Variability of the prior distributions on the random marker effects creates a variety of different Bayesian methods for genomic prediction. For examples, all the markers can be assumed to have distributions with different variances (Bayes A). Alternatively, some of the variances can be restricted to be zero (Bayes B). The Bayesian alphabet can be extended on and on. Consequently, some of the Bayesian methods are superior to gBLUP under some circumstances. For example, in situation that a trait is controlled by a few number of genes, Bayes B performs better than gBLUP. This talk presents the development of BLUP alphabet for the improvement of prediction accuracy. For examples, a newly developed compression BLUP (cBLUP) is superior to gBLUP for traits with low heritability controlled by many genes. In contrast, for a trait with high heritability controlled by a few genes, another newly developed SUPER BLUP (sBLUP) is superior to Bayesian methods. A final recommendation is given for a specific trait to select the optimum method according to genetic architecture defined by number of genes and heritability.
Back to: Genome Technology: Other Genome Methodology
<< Previous Poster | Next Poster >>