Assessing genomic prediction accuracy for Holstein sires using bootstrap aggregation sampling and leave-one-out cross...

Area of Science:

• Animal Genetics and Breeding
• Dairy Cattle Improvement
• Genomic Prediction

Background:

• Genomic selection has revolutionized dairy cattle breeding since 2009, enhancing selection response for economically important traits.
• Genomic Predicted Transmitting Ability (GPTA) guides critical breeding decisions for young dairy bulls.
• Current reliability (REL) values have limitations in identifying bulls with significant deviations between GPTA and actual daughter performance (DYD).

Purpose of the Study:

• To evaluate if current REL and bootstrap standard deviation (SD) can identify bulls with significant GPTA-DYD discrepancies.
• To identify factors within the reference population that contribute to inaccurate genomic predictions.
• To assess the utility of bootstrap aggregation sampling (bagging) with genomic BLUP (GBLUP) for improved prediction accuracy.

Main Methods:

• Applied bootstrap aggregation sampling (bagging) with GBLUP to predict GPTA for Holstein bulls.
• Utilized a reference population of older Holstein bulls with Daughter Yield Deviations (DYD) for 50 bootstrap samples.
• Performed leave-one-out cross-validation to assess prediction accuracy and identify influential reference population bulls.

Main Results:

• Bootstrap SD showed mild utility in identifying bulls with future performance deviating from early GPTA for protein yield and daughter pregnancy rate (DPR).
• Leave-one-out cross-validation identified influential reference population groups affecting protein yield predictions.
• The study highlights limitations of current REL and suggests alternative variability measures for more accurate genomic predictions.

Conclusions:

• Bootstrap SD and leave-one-out cross-validation offer potential improvements over current REL for identifying prediction inaccuracies in genomic selection.
• Understanding reference population structure is crucial for improving the reliability of genomic predictions in dairy cattle.
• Further research is needed to refine methods for detecting and mitigating inaccurate genomic predictions to optimize dairy breeding programs.