基因组选择技术在作物育种中的应用与进展OA

To comprehensively elucidate the current status and application potential of genomic selection(GS)in crop genetic improvement and breeding,a total of 253 publications from Web of Science and CNKI were retrieved by using"genomic selection","crop breeding","deep learning"and"statistical models"as keywords.This study systematically summarized research progress in theoretical foundations,methodological evolution,model optimization,application cases,and outstanding challenges of GS,and focused on how specific strategies,such as multi-omics integration and neural-network-based approaches,can improve the predictive accuracy of GS-assisted breeding.The results indicate that:GS has expanded from conventional phenotype-based selection theory to modern breeding practices that integrate high-throughput genomic data.Meanwhile,GS models have evolved from early linear statistical approaches centered on best linear unbiased prediction to Bayesian models,regularization methods,multi-trait prediction,genotype-by-environment(G×E)modeling frameworks,and machine-learning and deep-learning models,which have shown marked advantages for predicting complex traits.In addition,major crops such as soybean have become representative systems for applying GS in breeding,encompassing model training,optimizing of training populations,and modeling of G×E interaction.In the future,breakthroughs in GS will depend on multi-omics data integration,robust modeling of environment-genotype interactions,interpretable algorithms,multimodal deep-learning frameworks,and the construction of large-scale training datasets tailored to practical breeding programs.Overall,GS is becoming the key technological pillar of next-generation intelligent breeding systems and providing an important leverage point for achieving genetic gain and sustainable crop improvement under constraints of limited arable land and ongoing climate challenges.