Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Animal Genomics and Improvement Laboratory » Research » Publications at this Location » Publication #345801

Title: Use of causative variants and SNP weighting in a single-step GBLUP context

item FRAGOMENI, BRENO - University Of Georgia
item LOURENCO, DANIELA A. - University Of Georgia
item Tooker, Melvin
item Vanraden, Paul
item MISZTAL, IGNACY - University Of Georgia

Submitted to: World Congress of Genetics Applied in Livestock Production
Publication Type: Proceedings
Publication Acceptance Date: 2/7/2018
Publication Date: 2/7/2018
Citation: Fragomeni, B.O., Lourenco, D.L., Tooker, M.E., Van Raden, P.M., Misztal, I. 2018. Use of causative variants and SNP weighting in a single-step GBLUP context. World Congress of Genetics Applied in Livestock Production. Auckland, New Zealand, Feb. 11–16, Vol. Methods & Tools–Prediction 1, p. 140.

Interpretive Summary:

Technical Abstract: Much effort has been recently put into identifying causative quantitative trait nucleotides (QTN) in animal breeding, aiming genomic prediction. Among the genomic methods available, single-step GBLUP (ssGBLUP) became the choice because of its simplicity and potentially higher accuracy. When QTN are known, they need to be properly weighted so the accuracy can be maximized. The weighted ssGBLUP is still under development, and a proper weighing algorithm is needed. The objectives of this study were to investigate whether ssGBLUP is useful for genomic prediction when causative variants are known and to verify the impact of different SNP weighting in ssGBLUP compared to GBLUP. Analyses involved about 4M records for stature on 3M cows, and 4.7M animals in the pedigree. Genotypes were available for 27k sires for a regular 54k chip (BovineSNP50; Illumina), and with extra 17k sequence variants having largest effects, including causative variants. Direct genomic value (DGV) and genomic EBV (GEBV) were calculated using GBLUP and ssGBLUP with regular genomic relationship matrices (G). Later, G received weights calculated based on two different approaches: linear and nonlinear A. In the first, weights were calculated as SNP effect squared; whereas the latter was a fast version of BayesA that limits the changes in SNP weights to a small range. In GBLUP, the residuals were either homogeneous or heterogeneous. Reliability (R2) was assessed from regression of daughter deviations on GEBV or DGV for young sires with at least 10 daughters with records in the complete data. The lowest to highest R2 for DGV were by GBLUP with homogeneous residuals, GBLUP with heterogeneous residuals, and extracted from ssGBLUP. SNP weighting by nonlinear A increased R2 by up to 1.7% with homogeneous residuals and by up to 0.2% with heterogeneous residuals, compared to unweighted GBLUP. Linear weighting reduced accuracy in GBLUP and had no effect in ssGBLUP. Overall, adding 17k causative variants increased accuracy up to 0.6% in GBLUP, but had no impact in ssGBLUP. Reliability for DGV extracted from ssGBLUP was at least 0.6% more accurate than any DGV from GBLUP. Linear weighting is not helpful with small causative variants. Gains with SNP weighting in multistep (GBLUP or SNP BLUP) may be partly due to corrections in modeling issues associated with pseudo-observations.