W593 Tools for Genomic Analyses Using Single-Step Methodology

Date: Sunday, January 15, 2012
Time: 10:30 AM
Room: Sunset
Ignacy Misztal , University of Georgia, Athens, GA
Huiyu Wang , University of Georgia, Athens
Ignacio Aguilar , INIA, Uruguay
Andres Legarra , INRA, Castanet-Tolosan, France
William Muir , Purdue University, West Lafayette, IN
A single-step GBLUP (ssGBLUP) is a procedure that calculates GEBVs based on combined pedigree, genomic and phenotypic information. The procedure achieves these goals by blending traditional pedigree relationships with those derived from genetic markers. In practical applications, ssGBLUP exhibited superior accuracy with simplicity of application and inexpensive computing. Recently, ssGBLUP was expanded to 1) obtain approximate accuracies of GEBVs, 2) provide supplemental predictions for young animals based on a full-scale analysis, and 3) for genome-wide association analysis (GWAS). The last modifications allows for differential weightings of SNP. For GWAS, GEBVs are computed initially using a genomic relationship matrix (GRM) derived assuming equal weight per SNP. Then GEBVs are converted to SNP effects, which are used to estimate SNP weights. The weights are then applied to modify GRM, and the procedure is repeated iteratively. Efficiency of the method was examined using simulations with 15,800 subjects of which 1500 were genotyped. Thirty QTLs were simulated across genome and assumed heritability was 0.5. Comparisons included ssGBLUP and two Bayesian regression estimators, BayesA and BayesB. An accuracy of prediction of 0.85 was obtained by ssGBLUP after only two iterations, which was slightly higher than BayesA or BayesB. Power and precision for GWAS applications was evaluated by correlation between true QTL effects and the sum of n adjacent SNP effects, where n varied from 1-100. The highest correlations were achieved with n=40 and were 0.94 for ssGBLUP and 0.93 for BayesB. Computing time for ssGBLUP took about 2min while BayesB took about 4h. The ssGBLUP with marker weights is 2 orders of magnitude faster than the next best procedure, accurate, and easy to implement for GWAS applications.