Predicting Yield of Hybrid Rice Using Omics Data

Yield is the most important trait in rice. However, it is a low heritability trait with a heritability roughly equal to 10%. This means that the predictability (squared correlation between observed and predicted yields) using phenotypic value is only 10%. We now have genomic and other omic data, such as transcriptome, proteome, metabolome and phenome. These data can be used to predict yield possibly with an increased predictability. More importantly, we can use these data to predict yield of hybrid rice. Using an immortalized F₂ hybrid population (generated from intercrosses of recombinant inbred lines) and advanced statistical methods, e.g., best linear unbiased prediction (BLUP) and least absolute shrinkage and selection operator (LASSO), we demonstrated that genomic data can increase predictability up to 20%, transcriptomic data up to 23% and metabolomic data up to 35%. Using a training sample of 278 hybrids derived from the crosses of 210 recombinant inbred lines, we predicted the yield of all 210 x (210 - 1)/2 = 21945 potential crosses. The average yield of all crosses is 43.5g. The average yield of predicted top 10 crosses from genomic data is 49.44g, representing a 13.7% increase in yield. The average yield of predicted top 10 crosses from metabolomics data is 56.38g, representing a 29.6% increase in hybrid yield. Metabolomic prediction for low heritability traits seems to be a very promising technology, not only for hybrid rice but also for hybrid maize and other crops where hybrid breeding is the main tool for genetic improvement.