Advancements in population improvement methods, have greatly increased the number of inbred lines generated per breeding cycle. Consequently, the number of possible cross combinations between the inbreds becomes extremely large. As the number of combinations increases rapidly, their field evaluation becomes resource intensive and cumbersome. Thus, it is of great importance to the breeder, to identify the most promising hybrids for field evaluation through predictive models that can greatly accelerate breeding program and genetic gain.

    Crop breeding by prediction refers to assessing the performance of individual plants, clones, lines, testcrosses, hybrids, or populations via procedures that do not involve phenotyping the candidates themselves1. It has been started in the 1930s when superior double-cross maize hybrids were best predicted based on the mean performance of the four nonparental single crosses. Later, this method also became successful in predicting the performance of three-way crosses and synthetic populations.

    Different methods of prediction have been proposed by several researchers. They are mainly classified into two methods based (i) on phenotypic data and (ii) approaches exploiting phenotypic data along with molecular marker. The first group comprises hybrid prediction based on line per se performance (per se), general combining ability (GCA) and best linear unbiased prediction (BLUP). BLUP is a standard method for estimating random effects of a mixed model. It allows comparisons among genotypes developed from different breeding populations and evaluated in different sets of environments. The phenotype-based prediction has proven ineffective in predicting hybrid performance due to the prevalence of strong non-additive effects. Therefore, the focus in evaluation of hybrid performance has since shifted to genomic prediction (GP) which includes marker effect estimation-based approaches.

    Many prediction models are available for genomic selection (GS) viz., parametric methods – GBLUP, RRBLUP, Bayesian models, and LASSO and non-parametric methods such as machine learning - RKHS, random forest, and support vector machine. An attempt was made in maize to predict the breeding values of genotypes under drought stress using seven GS models. 240 maize subtropical lines phenotyped for drought at different environments and genotyped using 29,619 cured SNPs. Among several models, Bayes B was shown to have the highest prediction accuracy.

    Predictive breeding circumvents the phenotyping of the candidates themselves. Further, the use of genomic predictions over phenotype-based selection in breeding reduces the cost per cycle and the time required for variety development. Additionally, advanced GS models which include environmental covariates could be employed in predictive breeding programs to effectively reduce number of multiple-location trials.

References:

1BERNARDO, R., 2021, Predictive breeding in maize during the last 90 years. Crop Sci., 61(5):2872-2881.

2JENKINS, M.T., 1934, Methods of estimating the performance of double crosses in corn. J. Am. Soc. Agron., 26:199–204.

3SHIKHA, M., KANIKA, A., RAO, A.R., MALLIKARJUNA, M.G., GUPTA, H.S. AND NEPOLEAN, T., 2017, Genomic selection for drought tolerance using genome-wide SNPs in maize. Front. plant Sci., 8: 550.

4ZHAO, Y., METTE, M.F. AND REIF, J.C., 2015, Genomic selection in hybrid breeding. Plant Breed., 134(1):1-10.