Reducing computational cost of large-scale genomic evaluation by using indirect genomic prediction

Area of Science:

• Animal Genetics and Genomics
• Quantitative Genetics
• Statistical Genetics

Background:

• Annual genotyping of over 500,000 US Holsteins presents computational challenges for single-step genomic best linear unbiased prediction (ssGBLUP).
• Genomic prediction for young genotyped animals without progeny or phenotypes is computationally intensive.
• Indirect genomic predictions (IGP) are proposed to reduce computational costs for these animals.

Purpose of the Study:

• To calculate IGP for genotyped Holstein animals lacking phenotypes or progeny.
• To evaluate the reduction in computing time, bias, and accuracy of IGP compared to genomic (G)EBV.
• To determine an optimal number of genotyped animals for calculating SNP effects to balance accuracy and computational efficiency.

Main Methods:

• Compared IGP with G-EBV for 18 linear type traits in US Holsteins (2.3 million genotyped animals).
• Calculated SNP marker effects using GEBV from genotyped animals with progeny and/or phenotypes.
• Computed IGP for target genotyped animals (2014-2018) using ssGBLUP, excluding animals without progeny or phenotypes.

Main Results:

• High coefficients of determination (R² ≈ 0.96 for males, 0.95 for females) between GEBV and IGP across 18 traits.
• Randomly selecting 25K-35K genotyped animals to compute SNP effects yielded accurate and unbiased IGP, dramatically reducing computing time.
• Increased number of selected animals improved R² (accuracy) but also increased IGP inflation.

Conclusions:

• IGP is an accurate and unbiased method for genomic evaluation of young genotyped animals in large Holstein populations.
• Separating genomic evaluations into GEBV for animals with data and IGP for young animals offers a practical, cost-effective solution.
• This approach enables more frequent and less costly genomic evaluations, especially crucial for large genotyped populations.