Predicting hybrid performance in rice using genomic best linear unbiased prediction

Abstract

Genomic selection is an upgrading form of marker-assisted selection for quantitative traits, and it differs from the traditional marker-assisted selection in that markers in the entire genome are used to predict genetic values and the QTL detection step is skipped. Genomic selection holds the promise to be more efficient than the traditional marker-assisted selection for traits controlled by polygenes. Genomic selection for pure breed improvement is based on marker information and thus leads to cost-saving due to early selection before phenotypes are measured. When applied to hybrid breeding, genomic selection is anticipated to be even more efficient because genotypes of hybrids are predetermined by their inbred parents. Hybrid breeding has been an important tool to increase crop productivity. Here we proposed and applied an advanced method to predict hybrid performance, in which a subset of all potential hybrids is used as a training sample to predict trait values of all potential hybrids. The method is called genomic best linear unbiased prediction. The technology applied to hybrids is called genomic hybrid breeding. We used 278 randomly selected hybrids derived from 210 recombinant inbred lines of rice as a training sample and predicted all 21,945 potential hybrids. The average yield of top 100 selection shows a 16% increase compared with the average yield of all potential hybrids. The new strategy of marker-guided prediction of hybrid yields serves as a proof of concept for a new technology that may potentially revolutionize hybrid breeding.

Keywords: IMF2; hybrid rice; mixed model; restricted maximum likelihood; variance component analysis.

Fig. 1.

Goodness of fit (Upper) and predictability (Lower) of four traits plotted against model size, where the model size is determined by the number of variance components.

Fig. 2.

Effects of model size and sample size on the goodness of fit (Upper) and the predictability (Lower) of genomic prediction.

Fig. 3.

Average predicted genomic value of selected top crosses plotted against the number of crosses selected. The two dotted curves define the 95% confidence intervals of the mean predicted genomic value. The minimum value of the y axis for each trait is the average predicted genomic value for that trait. The plot is truncated at 500, and the total number of top crosses can run to 21,945 (all potential crosses).