Estimation of Genomic Breed Composition for Purebred and Crossbred Animals Using Sparsely Regularized Admixture...

Area of Science:

• Animal Genetics
• Statistical Genomics
• Bioinformatics

Background:

• Admixture models are standard for estimating genomic breed composition (GBC).
• Existing models often yield non-zero components for related breeds, inflating false-negative rates in purebred identification.
• This necessitates arbitrary lower cutoffs, reducing interpretability.

Purpose of the Study:

• To propose and evaluate three regularized admixture models for more accurate GBC estimation.
• To compare their performance against non-regularized models, particularly for purebred and composite animals.
• To assess the impact of different regularization penalties (L1, MCP, SCAD) on GBC estimation.

Main Methods:

• Development of three admixture models incorporating sparse regularization (L1, MCP, SCAD penalties).
• Application of models to estimate GBC in purebred and composite cattle (Brangus).
• Comparison of regularized models against a baseline non-regularized admixture model.

Main Results:

• Regularized models produced sparser GBC estimates, suppressing noise from genomic similarities.
• All three regularized models significantly reduced false-negative rates in purebred identification compared to the non-regularized model.
• Non-convex penalties (MCP, SCAD) outperformed the L1 norm penalty.
• All models underestimated GBC in composite Brangus cattle when non-ancestral breeds were included.

Conclusions:

• Sparse regularization enhances the parsimony, consistency, and interpretability of GBC estimates for purebred animals.
• Regularized admixture models offer improved accuracy for identifying purebreds.
• Application to composite or crossbred animals requires careful consideration due to potential underestimation.