Pangenomes can facilitate a deeper understanding of genome complexity. Using de novo phased long-read assemblies of eight representative goat breeds, we constructed a graph-based pangenome of goats (Capra hircus) and discovered 113-Mb autosomal novel sequences. Combining this multi-assembly pangenome with low-coverage PacBio HiFi sequences, we constructed a long-read structural variations (SVs) database containing 59,325 SV deletions, 84,910 SV insertions, and 24,954 other complex SV alleles. This resource allowed reliable graph-based genotyping from short reads of 79 wild and 1,148 worldwide domestic goats. Selection signal analysis of SV captured a novel immune-related domestication locus containing the galectin-9 gene and extra copies of the ruminant-specific galectin-9-like genes (LGALS9L), which have high tissue specificity. A segmental duplication in domestic goats generates three additional LGALS9L copies. Ancient goat genome sequences show a gradual increase in frequency of this duplication from the Neolithic to the present. Two other newly detected SVs also have higher selection signals than adjacent SNPs, a truncated-LINE1 deletion in EDAR2 associated with cashmere production and a VNTR-related insertion in PAPSS2 linked to high-altitude adaptation. In summary, the multi-assembly goat pangenome and long-read SV database facilitates detecting complex variations that are important in evolution and selection.

Goats are essential to the dairy industry in Shaanxi, China, with udder traits playing a critical role in determining milk production and economic value for breeding programs. However, the direct measurement of these traits in dairy goats is challenging and resource-intensive. This study leveraged genotyping imputation to explore the genetic parameters and architecture of udder traits and assess the efficiency of genomic prediction methods. Using data from 635 Saanen dairy goats, genotyped for over 14,717,075 SNP markers and phenotyped for three udder traits, heritability was 0.16 for udder width, 0.32 for udder depth, and 0.13 for teat spacing, with genetic correlations of 0.79, 0.70, and 0.45 observed among the traits. Genome-wide association studies (GWAS) revealed four candidate genes with selection signatures linked to udder traits. Predictive models, including GBLUP, kernel ridge regression (KRR), and Adaboost.RT, were evaluated for genomic estimated breeding value (GEBV) prediction. Machine learning models (KRR and Adaboost.RT) outperformed GBLUP by 20% and 11% in predictive accuracy, showing superior stability and reliability. These results underscore the potential of machine learning approaches to enhance genomic prediction accuracy in dairy goats, providing valuable insights that could contribute to improvements in animal health, productivity, and economic outcomes within the dairy goat industry.