Abstract:

The extensive application of genomic selection has significantly accelerated genetic advancements in livestock and poultry. With the commercialization of livestock and poultry chips and the continuous reduction of sequencing costs, the available genomic information for livestock and poultry has become increasingly abundant. Many challenges have arisen in genomic selection, such as the number of genotypic markers far exceeds the number of samples with phenotype data, and the relationships between genomic information have become more complex. These problems greatly restrict the use of traditional evaluation models such as best linear unbiased prediction (BLUP) and Bayes. Machine learning algorithms, which do not rely on predetermined equation models, have demonstrated superior capability in handling nonlinear relationships. Machine learning methods can offer solutions to the aforementioned challenges, thus they are gradually being applied in genomic selection. This paper reviewed the developmental of genomic selection, elucidated the principles of several commonly used machine learning algorithms. Furthermore, the current status and implementation methods of machine learning in livestock and poultry genomic selection were summerized. Finally, the challenges faced by machine learning in livestock and poultry breeding, as well as its development prospects were discussed.

Key words: genomic selection, livestock and poultry, machine learning, algorithm, model