Using genetic regressions to account for genomic preselection effects in MACE


  • Peter G Sullivan Canadian Dairy Network
  • Esa Mäntysaari Natural Resources Institute (Luke)
  • Gerben de Jong CRV u.a.
  • Simone Savoia Interbull Centre


National genomic evaluation systems use foreign sire evaluations from MACE as phenotypic information, combined with genotypes to generate national GEBV.  The national GEBV computed from MACE cannot be used as input data for MACE without double-counting genomic information.  To avoid this double-counting, Interbull requires that national EBV provided as input for MACE must be computed without genotypes, even though this leads to known biases in MACE results.  The biases are due to sire pre-selection effects based on genotypes being partially treated as effects of the sires’ mates and the herd environments of the sires’ progeny when genotypes are excluded from the national evaluations.  This causes an underprediction of genetic levels for genomically pre-selected sires and of estimated genetic trends, for both the input data and the output results of MACE.  The current MACE model was therefore expanded by adding estimates of GPS effects, and with corresponding modifications to the yearly means of estimated breeding values for genomically pre-selected sires.  Segmented genetic regressions were used to estimate evolving international trends in pre-selection effects since 2009, and genetic grouping was used to include pre-selection estimates in the genetic evaluations of genomically pre-selected sires.  Data simulation was used to validate the expanded MACE model, under a scenario where GPS effects are fully included, and GPS biases are thus zero in the national input data.  The new MACE model properly separated within-family from between-family pre-selection effects in the simulated data and effectively removed pre-selection biases observed under the current MACE model.  Estimated pre-selection effects were relatively small from official data but are expected to increase as national models used to generate MACE input data will be updated to reduce genomic pre-selection biases in the future.  Further improvements are also being planned for the new MACE model, to account for expected genetic variance reductions with the elevated means for GPS bulls due to selection.