Animal Molecular Breeding, 2025, Vol.15, No.2, 49-59 http://animalscipublisher.com/index.php/amb 52 developed. It can help researchers understand more clearly how these complex models make judgments and extract biologically significant explanations. 4.2 AI applications in genotype-phenotype prediction At present, many kinds of machine learning algorithms have been used to predict the carcass characteristics, growth rate and health status of livestock. The effects of these AI models are sometimes even better than those of traditional linear models (Liang et al., 2020; Srivastava et al., 2021; Chafai et al., 2023). Morota et al. (2022) and Dórea and Menezes (2024) demonstrated that AI is also frequently employed in the collection of phenotypic data. Researchers can monitor the growth, behavior and health of animals in real time through computer vision technology and wearable sensors, and can collect high-quality data quickly and accurately. Ferreira et al. (2024) found that AI can also integrate various data from different sources, such as sensor information and images, to enhance the accuracy of predictions and facilitate the early detection of health issues. 4.3 Advantages of AI in handling complex data sets Machine learning and deep learning models can well capture various relationships in data, whether they are simple linear relationships or more complex nonlinear patterns. They can extract useful genetic information from very large SNP data and also cope with some non-additive genetic influences (Liang et al., 2020; Srivastava et al., 2021; Hay, 2024). Ferreira et al. (2024) demonstrated that AI-driven data fusion technology can integrate data from different sources, enabling more accurate and timely analysis. AI also makes it possible to collect large-scale phenotypic data of animals without disturbing them, as it can automatically complete many links, such as checking data quality and selecting key features, improving the efficiency and sustainability of livestock breeding (Morota et al., 2022; Dórea and Menezes, 2024; Spangler, 2024). 5 Framework for Integrating AI with Genomic Selection 5.1 AI-enhanced genomic prediction pipelines New advancements in AI make it easier for us to identify complex genetic structures that are difficult to understand with traditional linear models, such as nonlinear relationships and interactions between genes. Adding AI methods to the genome selection process can combine a large amount of genomic and phenotypic data and improve the prediction accuracy of breeding values. With deep learning and other advanced machine learning methods, AI models can handle ultra-large-scale and multi-dimensional data, making genomic breeding faster and more accurate. This method is particularly useful for traits that are controlled by many genes and are easily affected by the environment. It is very likely to break the traditional breeding methods and accelerate the improvement speed of domestic animal breeds such as goats (Figure 2) (Bhat et al., 2023). 5.2 Data sources and preprocessing strategies Efficient AI-driven genome selection (GS) relies on diverse and high-quality data resources, such as high-density SNP chips, whole-genome sequencing data, and rich phenotypic records. Take goats as an example. The 52K SNP chip developed by the International Goat Genome Consortium has been widely used in genome-wide association studies (GWAS) and GS studies (Rupp et al., 2016). Before modeling, data preprocessing is very crucial, usually including the quality control of genotype data, filling in missing values, standardization of phenotypic data, and integration of multi-omics information. Simulation studies show that using medium-density SNP panels (such as 45K SNPs) combined with a moderate-sized reference population (approximately 1 500) can effectively improve the prediction accuracy of goat GEBV. The accuracy of GEBV is also related to the heritability of the trait and the number of RAMS in the reference population. For traits with medium heritability, if the reference population size is large, the prediction effect will be better (Yan et al., 2022). 5.3 Comparison with traditional BLUP and GBLUP models BLUP uses family and phenotypic data, while GBLUP adds DNA information on this basis. GBLUP is particularly suitable when there is not much phenotypic data, and its predictive effect will be more accurate. The later emerged single-step GBLUP (ssGBLUP) integrates genotype, phenotype and lineage data all together, and its predictive accuracy is slightly higher than that of GBLUP. The research results of Yan et al. (2022) on dairy goats
RkJQdWJsaXNoZXIy MjQ4ODYzNA==