Bias due to selective genotyping in genomic prediction using H-BLUP

TY - CONF

T1 - Bias due to selective genotyping in genomic prediction using H-BLUP

AU - Wang, Lei

AU - Madsen, Per

AU - Sapp, Robyn

AU - Henshall, John

AU - Hawken, Rachel

AU - Alemu, Setegn Worku

AU - Sørensen, Anders Christian

AU - Jensen, Just

PY - 2016/6/12

Y1 - 2016/6/12

N2 - H-BLUP uses a variance-covariance structure based on a combined relationship matrix (H), which augments a pedigree-based relationship matrix (A) with a genomic relationship matrix (G) for genotyped individuals. In practice, often only preselected individuals are genotyped and this selective genotyping may generate bias in estimated variance components (VC) and predicted breeding values (EBVs). Most current applications of H-BLUP use VC from traditional animal models, and use observed allele frequencies computed from genotyped individuals only to construct G, since base population allele frequencies are usually not available. However, such VC doesn’t take ongoing selection based on genomic information into account. The aim of this study is to investigate the effects of selective genotyping and of different strategies to compute G on the accuracy and potential bias in VC estimates by using H-matrix, and in EBVs from H-BLUP. We simulated a population similar to a selection line of broilers with overlapping selection rounds. In the simulation, there were 20 rounds of BLUP selection after base population, followed by 20 rounds of H-BLUP selection. Results showed that bias in VC estimates using H-matrix increased when the proportion of genotyping best individuals increased from 10% to 30%, and EBVs of genotyped animals were also biased upwards in H-BLUP. In contrast, when genotyping was random, much less bias in VC estimates and EBVs were observed. Using VC estimated from traditional animal models in genetic evaluation by H-BLUP reduced the bias in EBVs, but did not remove it completely. Using base population allele frequencies to construct G did not improve VC estimation. We conclude that the bias in VC estimates from using H matrix is due to selective genotyping, and using base population allele frequency would not help remove the bias.

AB - H-BLUP uses a variance-covariance structure based on a combined relationship matrix (H), which augments a pedigree-based relationship matrix (A) with a genomic relationship matrix (G) for genotyped individuals. In practice, often only preselected individuals are genotyped and this selective genotyping may generate bias in estimated variance components (VC) and predicted breeding values (EBVs). Most current applications of H-BLUP use VC from traditional animal models, and use observed allele frequencies computed from genotyped individuals only to construct G, since base population allele frequencies are usually not available. However, such VC doesn’t take ongoing selection based on genomic information into account. The aim of this study is to investigate the effects of selective genotyping and of different strategies to compute G on the accuracy and potential bias in VC estimates by using H-matrix, and in EBVs from H-BLUP. We simulated a population similar to a selection line of broilers with overlapping selection rounds. In the simulation, there were 20 rounds of BLUP selection after base population, followed by 20 rounds of H-BLUP selection. Results showed that bias in VC estimates using H-matrix increased when the proportion of genotyping best individuals increased from 10% to 30%, and EBVs of genotyped animals were also biased upwards in H-BLUP. In contrast, when genotyping was random, much less bias in VC estimates and EBVs were observed. Using VC estimated from traditional animal models in genetic evaluation by H-BLUP reduced the bias in EBVs, but did not remove it completely. Using base population allele frequencies to construct G did not improve VC estimation. We conclude that the bias in VC estimates from using H matrix is due to selective genotyping, and using base population allele frequency would not help remove the bias.

M3 - Poster

T2 - 5th International Conference on Quantitative Genetics

Y2 - 12 June 2016 through 17 July 2016

ER -