Using Metafounders reduces bias in Single Step GBLUP genomic evaluations

LEGARRA, A.; VITEZICA, Z.G.; GARCÍA BACCINO, C.A.; CANTET, R.J.C.

Madison, Wisconsin

Congreso; 5th International Conference on Quantitative Genetics; 2016

Combining genomic and pedigree relationships requires the same base populations, that is, same average breeding values and genetic variance at the base. This is difficult to do when genotyped animals are younger or selected. To make base populations identical we consider each ancestral population as a finite-sized pool of gametes called metafounder. A metafounder is a pseudo-individual included as founder of the pedigree and similar to an ?unknown parent group?. The use of metafounders enables the modification of pedigree relationships to match genomic relationships. The latter are computed using the -101 coding for genotypes, then the relationship of the metafounder (gamma) is estimated using pedigree and markers. Gamma is proportional to the (co)variance of allelic frequencies in the pedigree founders. Allele frequencies were estimated using least squares (LS), generalized least squares (GLS) or maximum likelihood (ML).We carried out a simulation study (20 replicates) to explore the possibility of including metafounders in Single Step GBLUP (SSGBLUP). Data was simulated with QMSim mimicking a dairy cattle selection scheme. Ten recent generations were simulated coming from 100 historical generations including 40,000 segregating markers and 1,500 segregating QTLs. There were 28,800 individuals in the pedigree, 6,038 individuals with genotypes and only the females had phenotypic records. Generation 10 had no phenotype and was used for checking accuracy and bias.Estimates of gamma were accurate and unbiased if GLS or ML were used (0.40 and 0.39, respectively; 0.40 was the true value) but overestimated with LS (0.43). There was an important gain in terms of bias when including a metafounder in SSGBLUP (the regression coefficient of true on estimated breeding values was 0.82 for regular SSGBLUP and 0.94 when including the metafounder). There was also a slight gain in accuracy (0.74 vs 0.72). Using metafounders also produced more accurate variance components estimations.