A Proximal Multi-Objective Optimization Method for Incorporation of Polygenic Breeding Values in Genomic Prediction.

Area of Science:

• Quantitative genetics
• Genomic evaluation
• Bioinformatics

Background:

• Traditional quantitative genetics relies on pedigrees for breeding values.
• High-throughput sequencing enables empirical relationship estimation.
• Single-step genomic BLUP is computationally intensive due to complex pedigree data.

Purpose of the Study:

• Introduce DYpREG, a novel method for integrating polygenic and genomic information.
• Develop a computationally efficient approach for genomic evaluation.
• Enhance prediction accuracy in animal breeding.

Main Methods:

• DYpREG utilizes a splitting algorithm with proximal operators.
• Accommodates multiple loss and regularization functions (LASSO, RR, EN).
• Incorporates pre-calculated breeding values, avoiding extensive relationship matrix computations.

Main Results:

• DYpREG with Ridge Regression (RR) regularizer showed superior predictive performance.
• Achieved notable Mean Squared Error (MSE) reductions in mice (1.5%, 24.77%) and pigs (2.8%, 2.1%).
• Outperformed methods lacking polygenic information and showed improvements over Bayesian RKHS and BayesC.

Conclusions:

• DYpREG effectively integrates polygenic information into genomic evaluations.
• Offers a computationally efficient and accurate alternative for breeding value prediction.
• Demonstrates significant potential for enhancing genomic selection strategies.