Estimation of heritabilities and genetic correlations by time slices using predictivity in large genomic models

Abstract

Under genomic selection, genetic parameters may change rapidly from generation to generation. Unless genetic parameters used for a selection index are current, the expected genetic gain may be unrealistic, possibly with a decline for antagonistic traits. Existing methods for parameter estimation are computationally unfeasible with large genomic data. We present formulas for estimating heritabilities and genetic correlations applicable for large models with any number of genotyped individuals. Heritabilities are calculated by combining 2 formulas for genomic accuracies: one that relies on predictivity and another that depends on the number of independent chromosome segments. Genetic correlations are calculated from predictivities across traits. Both formulas include approximate standard errors. The formula for heritabilities was evaluated based on information for 4 commercial data sets extracted from published studies. Calculated heritabilities were close to those used initially, except for a much lower new heritability for one single case; that heritability was later confirmed as valid. Formulas for genetic correlations were tested with simulated data and 1,000 replicates. The formula-based estimates were always close to the values assumed for simulation. Standard errors were high with 1,000 validation animals but small with 10,000. The proposed formulas can be used routinely as a check on the evaluation system whenever the number of validation individuals is large enough. If excessive changes are detected from generation to generation, the selection index can be modified appropriately.