Ridge, Lasso and Bayesian additive-dominance genomic models.

Area of Science:

• Quantitative genetics
• Animal breeding
• Genomic selection

Background:

• Genome-wide selection (GWS) requires robust statistical models.
• Additive-dominance models are less explored than additive models in GWS.
• Current GWS methods need evaluation for additive-dominance effects.

Purpose of the Study:

• Compare 10 additive-dominance predictive models.
• Evaluate Bayesian, Lasso, and Ridge regression approaches.
• Decompose genomic heritability and accuracy by information source (LD, CS, PR).

Main Methods:

• Simulation study with varying heritability and genetic architectures.
• Fitted 10 additive-dominance models using Bayesian, Lasso, and Ridge.
• Assessed prediction accuracy for genomic breeding and total genotypic values.
• Decomposed heritability and accuracy into linkage disequilibrium (LD), co-segregation (CS), and pedigree relationships (PR).

Main Results:

• G-REML/G-BLUP and modified Bayesian/Lasso (BayesA*B*/t-BLASSO) performed best across scenarios.
• BayesA*B*-type methods better recovered the additive-dominance variance ratio.
• Information importance for accuracy: LD > CS ≈ PR not captured by markers.

Conclusions:

• G-BLUP, BAYESA*B* (-2,8), and BAYESA*B* (4,6) were the top-performing models.
• These models accurately predict genomic breeding and total genotypic values.
• Models are suitable for estimating additive and dominance effects in genomic models.