327 Practical problems and solutions using unknown parent groups in combined commercial pig sub-lines. (7th December 2018)

• Record Type: Journal Article
• Authors: Pocrnic, I; Lourenco, D; Tsuruta, S; Chen, C; Misztal, I
• Is Part Of: Journal of animal science. Volume 96(2018)Supplement 3
• Page Start: 124
• Page End: 125
• Publication Date: 2018-12-07
• Subjects: genetic groups -- genomic selection -- single-step genomic BLUP; Livestock -- Periodicals; Livestock; Electronic journals; Periodicals; 636.005
• DOI: 10.1093/jas/sky404.274
• Languages: English
• ISSNs: 0021-8812

Abstract

Abstract: Swine breeding companies sometimes combine two or more sub-lines of the same breed into one large line that is afterwards used for genomic selection. Differences between sub-lines can be modeled by unknown parent groups (UPG). Two such large lines, L1 and L2, each consisting of two or more sub-lines, experienced different issues when single-step genomic BLUP (ssGBLUP) was used. In line L1, UPG solutions and genomic EBV (GEBV) strongly depended on small changes to blending parameters, for pedigree (A ) and genomic (G ) relationship matrices in ssGBLUP (beta and alpha in blupf90 software). This issue was traced to weak connections between the sub-lines and subsequently high estimation errors of UPG functions. Changing UPG effects from fixed to random resolved the issue. Over the time, as the sub-lines become more connected, the issue would be resolved automatically. In the case of line L2, UPG solutions and GEBV were stable under different blending parameters, but small additions of data sometimes caused analyses to terminate abnormally; analyses ran normally after removing older genotypes. This issue was traced to correlations of off-diagonal elements of G and A fluctuations around 0.5. Such a correlation usually ranges from 0.55 to 0.75, and default checks for genomic relationships in the blupf90 family of programs will stop computations if the value is below the threshold. Lowering the threshold resolved the problem. Alternatively, the problem could have been resolved by removing the oldest genotypes. Commercial genomic selection can experience a number of problems due to a highly specific data structure, and intricate knowledge of modeling and genomics is needed to solve those problems.