Technical note: An R package for fitting Bayesian regularized neural networks with applications in animal breeding

Area of Science:

• Quantitative genetics
• Statistical genomics
• Bioinformatics

Background:

• Accurate prediction of genetic values for complex traits is crucial in animal and plant breeding.
• Existing statistical models often rely on molecular markers and pedigree data.
• Bayesian regularized neural networks (BRNN) show promise for genomic prediction.

Purpose of the Study:

• To introduce the R package 'brnn' for implementing BRNN models.
• To extend BRNN models to include additive and dominance genetic effects.
• To demonstrate the package's utility with real-world examples.

Main Methods:

• Implementation of BRNN models within an R package.
• Extension of BRNN to account for additive and dominance genetic effects.
• Utilizing multicore architectures with OpenMP for parallel computation.

Main Results:

• The 'brnn' package provides an efficient implementation of BRNN for genomic prediction.
• The package successfully incorporates both additive and dominance effects.
• Parallel computing significantly speeds up model computations.

Conclusions:

• The 'brnn' R package offers a powerful and efficient tool for genome-enabled prediction.
• The inclusion of additive and dominance effects enhances prediction accuracy for complex traits.
• The package facilitates advanced statistical modeling in quantitative genetics.