Advancements in population improvement methods and Double Haploid techniques, have greatly increased the number of inbred lines generated per cycle within each heterotic group. Consequently, it became unfeasible to phenotypically evaluate performance of all single-cross hybrids due to the overwhelming number of pairwise combinations of inbred lines2. Hence, successful prediction of hybrid performance of untested hybrids based on the data of tested hybrids is an attractive alternative to expensive field testing for the identification of promising hybrids and can greatly accelerate hybrid breeding programs.

The Phenotype based prediction (inbred lines per se, GCA and BLUP) has proven ineffective in predicting hybrid performance due to the prevalence of strong dominance effects. Therefore, the focus in evaluation of hybrid performance has since shifted to genomic prediction (GP). In GP, a precisely phenotyped and genome-wide marker genotyped training population is used to develop a prediction model that predicts the effect of each marker corresponding to a trait. Later GP models are used to estimate breeding values referred to as genomic estimated breeding value of the individuals in the breeding population based on the genotypic data1.

Many prediction models are available for genomic prediction viz., parametric methods – GBLUP, RRBLUP, Bayesian models, and LASSO and non-parametric methods such as machine learning - RKHS, random forest, and support vector machine. In an attempt to predict the GCA and SCA effects of newly developed inbred lines for grain yield predictions were made using GP models. The predictive abilities of the hybrids ranged from 0.59 to 0.81 considering only additive effects, which increased to 0.64 to 0.86 when both additive and non-additive effects were incorporated into the model3.

GP offers not only the opportunity to predict the performance of new hybrids and the GCA of new inbred lines based on the genotypic information, but also the total breeding cost could be reduced dramatically by phenotyping fewer multiple-location trials. Additionally, there is a need for a GS model that includes the environment as a covariate to enhance prediction accuracy.

References:

1. BERNARDO, R., 2014, Genomewide selection when major genes are known. Crop Sci., 54(1):68-75.

2. SCHRAG, T. A., FRISCH, M., DHILLON, B. S. AND MELCHINGER, A. E., 2009, Marker-based prediction of hybrid performance in maize single-crosses involving doubled haploids. Maydica, 54:353-362.

3. ZHANG, A., PEREZ-RODRIGUEZ, P., SAN VICENTE, F., PALACIOS-ROJAS, N., DHLIWAYO, T., LIU, Y., CUI, Z., GUAN, Y., WANG, H., ZHENG, H. AND OLSEN, M., 2022, Genomic prediction of the performance of hybrids and the combining abilities for line by tester trials in maize. Crop Sci., 10(1):109-116.