Performance Comparison of Genomic Best Linear Unbiased Prediction and Four Machine Learning Models for Estimating...

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Canine Genomics

Background:

• Guide dog success is influenced by health and behavioral traits.
• Accurate prediction of genomic breeding values (gEBVs) is crucial for selective breeding.
• Various genomic prediction models exist, but their comparative efficiency for canine traits is not fully understood.

Purpose of the Study:

• To evaluate the performance of five genomic prediction models (GBLUP, RF, SVM, XGB, MLP) for predicting gEBVs in guide dogs.
• To assess the impact of trait heritability, case counts, and SNP density on model performance.
• To identify the most efficient model for genomic prediction in canine breeding programs.

Main Methods:

• Utilized phenotypic and genomic data from a population of 2050 guide dogs (German Shepherds, Golden Retrievers, Labrador Retrievers, and crosses).
• Assessed five genomic prediction models: GBLUP, Random Forest, SVM, XGBoost, and MLP.
• Analyzed model performance across four traits (anodontia, distichiasis, oral papillomatosis, distraction) with varying heritabilities and case counts, and across different SNP marker densities.

Main Results:

• All tested models demonstrated similar predictive performance across varying heritabilities, case counts, and SNP densities.
• Genomic Best Linear Unbiased Prediction (GBLUP) was the most efficient model due to its lack of requirement for parameter optimization.
• Distichiasis exhibited the highest predictive accuracy, followed by anodontia and distraction, with oral papillomatosis showing the lowest accuracy, correlating with their respective heritabilities.

Conclusions:

• Lower-density SNP datasets are sufficient for constructing accurate gEBVs, reducing the need for high-cost genotyping.
• Simpler models like GBLUP are adequate for genomic prediction in canine breeding programs, offering efficiency and ease of use.
• Standardized phenotypic assessments and well-constructed reference populations are vital for optimizing genomic selection in canine breeding.