Haseman-Elston回归与QR分解结合的全基因组选择新方法OA

The rapid development of high-throughput sequencing and phenotyping technologies has generated increasing large-scale genotype and phenotype datasets.The conventional genomic selection(GS)algorithm struggle to handle such data,and computational efficiency has become a key concern in genomic selection.In order to address this problem,this study pro-poses a novel computationally efficient Haseman-Elston regression+QR decomposition genomic selection(HQGS)method.First,Haseman-Elston regression is used to estimate genetic variance,followed by orthogonal triangular decomposition(QR decompo-sition)of the core population to obtain an approximate solution for high-dimensional matrices,which greatly improves computa-tional efficiency.In simulated data analysis,different levels of core population size,number of markers,true heritability,and training population size were set to evaluate HQGS,genomic best linear unbiased prediction(GBLUP),random forest(RF),and support vector machine(SVM).In terms of prediction accuracy,HQGS and GBLUP were similar in most cases,both significantly outperforming RF in all cases and outperforming SVM in most cases.In terms of computational efficiency,HQGS significantly outperformed GBLUP,RF,and SVM,with GBLUP being the least efficient.In real data analysis,for 14 traits in wheat population 1,HQGS showed significantly better prediction accuracy than GBLUP for 3 traits,was similar to GBLUP for 8 traits,signifi-cantly outperformed RF for 6 traits,was similar to RF for 5 traits,significantly outperformed SVM for 5 traits,and was similar to SVM for 5 traits.For yield data under four environments in wheat population 2,HQGS outperformed GBLUP in three environ-ments,outperformed RF in one environment,was similar to RF in two environments,outperformed SVM in two environments,and was similar to SVM in one environment.This study presents a novel computationally efficient genomic selection method that effectively avoids the inversion of large genetic relationship matrices,significantly improving computational efficiency while maintaining prediction accuracy,providing a more efficient and reliable new approach for handling large datasets.