Issues in commercial application of single-step genomic BLUP for genetic evaluation in American Angus Daniela Lourenco S. Tsuruta, B. Fragomeni, Y. Masuda, I. Pocrnic, I. Aguilar, J.K. Bertrand, D. Moser, I. Misztal University of Georgia July, 2016
ssGBLUP theory vs. practice Theory • Studies on several livestock species • Simple Practice • Full capability • Implementation for American Angus • Challenges and problems 2
Angus Data • 220,000 genotyped animals American Angus 250,000 • 220,000 9.7M pedigree 200,000 # Genotyped Animals 152,000 • BW, WW, PWG 150,000 132,000 112,000 • SC 100,000 82,000 • CE 50,000 • Docility 0 • Heifer pregnancy Number of Genotypes • Yearling height 07 2014 01 2015 07 2015 10 2015 07 2016 • Mature weight, height • Carcass weight, marbling, ribeye area, fat thickness • Dry matter intake 3
ssGBLUP - lots of genotyped animals Growth Traits – 82k −1 G APY 0.99 0.99 0.99 Correlation (GEBV,GEBV_APY) 0.97 CORE invert NON-CORE Misztal et al., 2014 5k 10k 15k 20k CORE animals randomly sampled from genotyped population 4
ssGBLUP - lots of genotyped animals • How to choose number of core animals? Pocrnic et al., 2016 • Ne, Me, ESM, Eigen of G Misztal, 2016 • Limited dimensionality AAA – 82k AAA - 82K 1.00 14555 COR (GEBV,GEBV_APY) Number of Eigenvalues 15000 0.98 10605 14555, 0.99 10605, 0.99 0.96 10000 6166 6166, 0.98 0.94 3654 5000 3654, 0.96 0.92 0 0.90 90 95 98 99 0 2000 4000 6000 8000 10000 12000 14000 16000 % of Variance NUMBER OF EIGENVALUES 5
ssGBLUP – Set core animals for evaluation • 220,000 genotyped animals • 99% eigen 19,000 • Core = 19,000 high accuracy CORE invert • Keep core constant 1 year • Add extra core NON-CORE • Genomic set up = 30min 201,000 6
Growth Traits with external EBV • External EBV from ~10k Red Angus • 220,000 genotyped animals • 19,000 core • E = external • 201,000 non-core • I = internal • • T = PEV for E 9.7M pedigree • 7.4M BW Adapted from Legarra et al., 2007 Computing time • 8.1M WW BLUP = 8h • 4M PWG ssGBLUP = 12h • Maternal effect for BW & WW 7
Calving Ease is categorical • 220,000 genotyped animals • 2-trait BW-CE linear-threshold • 19,000 core • BLUP = 12h • 201,000 non-core • ssGBLUP = 4.5 days • 8.7M Pedigree • 1.4M CE • 91% easy • 9% difficult 8
Calving Ease is categorical 𝐈 −𝟐 = 𝐁 −𝟐 + 𝟏 𝟏 • Cblup90iod2: 2 nested rounds (α𝐇 + β𝐁 𝟑𝟑 ) −𝟐 – 𝐁 𝟑𝟑 −𝟐 𝟏 Description of parameters correlation with Scenario rounds hours pcg rounds alpha beta genomic traditional 40 - 60 12 - genomic 40 0.9 0.1 488 108 - 1 100 0.9 0.1 81 43 0.999 2 100 0.85 0.15 62 32 0.999 3 200 0.9 0.1 24 25 0.999 4 200 0.85 0.15 19 19 0.999 Working on OMP – 30% faster 9
Interim GEBV • SNP effect • Weekly evaluation • New genotypes daily d i = SNP weight = I • 220,000 genotyped animals • GEBV I = 𝐚 𝑣 • 19,000 core • 201,000 non-core • COR (GEBV I_CORE ,GEBV I_50k ) = 0.98 10
Accuracy of GEBV • GEBV published with accuracy • Large datasets • Measurement of precision • Impossible to invert Traditional 𝑞 are approximated 𝑠 and 𝑒 𝑗 • 𝑒 𝑗 (Misztal and Wiggans, 1988) • Accuracy = 1 - 𝑀𝐼𝑇 -1 𝑗𝑗 = 1 𝑀𝐼𝑇 𝑣𝑣 Diag(C ZZ+ ) = PEV 𝑠 + 𝑒 𝑗 𝑞 ) (λ + 𝑒 𝑗 11
Accuracy of GEBV Genomic 0 0 𝑗𝑗 = 1 Z′Z+ λA−1+ λ 𝑀𝐼𝑇 𝑣𝑣 𝑞 + 𝑒 𝑗 −1 𝑠 + 𝑒 𝑗 G−1−A 22 ) (λ + 𝑒 𝑗 0 𝑞 𝑠 𝑒 𝑗 𝑒 𝑗 𝑒 𝑗 ? How to approximate 𝑒 𝑗 12
Accuracy of GEBV • Approximation 1 • Approximation 2 𝑠 and 𝑒 𝑗 𝑞 • h 2 • 𝑒 𝑗 • # effective SNP • G • Average contribution from G • Minimum relationship ( δ ) • Accuracy of PA • Accuracy of PA 𝑠 + 𝑒 𝑗 = 1 + ℎ 2 ∗ 𝐹𝑇𝑁 ∗ 𝐻 − 𝐵 22 ∗ 𝐵𝑑𝑑 𝑄𝐵 𝑞 ) 𝑗≠𝑘 𝐵𝑑𝑑 𝑄𝐵 𝑗≠𝑘 𝑗𝑘 (𝑒 𝑗 𝑒 𝑗 = λ 𝑒 𝑗 𝑑𝑝𝑣𝑜𝑢(𝜀) +≈ 𝑑𝑝𝑣𝑜𝑢(𝜀) 13
Accuracy of GEBV Approximation 2 Approximation 1 0’12” 3’24” 14
Considerations Several problems and challenges All solved - multi-trait, categorical, maternal - external info - interim GEBV - accuracy of GEBV ssGBLUP ready for national Beef cattle evaluation - Angus in 2016 15
Acknowledgements 16
Recommend
More recommend
