Multitrait genomic prediction method in approximate genome-based kernel model

In order to cultivate excellent varieties, breeders need to evaluate multiple traits simultaneously. In this study, we developed an efficient large-scale multitrait genomic prediction method in approximate genome-based kernel model (MT-RHPK). The results of our simulation study showed that with similar or better predictive accuracy, MT-RHPK excels multitrait genomic best linear unbiased predictor (MT-GBLUP) significantly in computational time. Comparing MT-RHPK with single-trait GBLUP (ST-GBLUP), we found that when genetic correlation coefficients between traits were positive, the former demonstrated better predictive accuracy for low-heritability trait and similar predictive accuracy for high-heritability trait, and when genetic correlation coefficients between traits were negative, the former demonstrated better or similar predictive accuracy for low-heritability trait, but was outperformed for high-heritability trait in most cases. In 14 paired traits of bread wheat and rice datasets, the predictive accuracies of MT-RHPK, MT-GBLUP, and ST-GBLUP were similar in most cases. However, when biomass and maturity had high positive genetic correlation (0.766±0.004), MT-RHPK and MT-GBLUP demonstrated better predictive accuracy for maturity, and when biomass and glaucousness had high negative genetic correlation (-0.667±0.068), MT-RHPK and MT-GBLUP were outperformed for glaucousness. In general, MT-RHPK is a practical and efficient tool to perform simultaneous improvement of multiple traits in the large-scale genomic era.