Reproducing Kernels: Predicting Grain Carotenoid Traits For a Maize Breeding Program
Mary LaPorte, University of California, Davis
Vitamin A deficiency is a life-threatening issue for pregnant women and children that can lead to blindness, morbidity, and mortality. Humans cannot synthesize vitamin A — it must be consumed through the diet. Dietary deficiency can be alleviated by increasing the intake of biochemical compounds, known as provitamin A carotenoids, that are converted to vitamin A after ingestion into human and animal systems. In regions where maize (Zea mays ssp. mays L.) constitutes a high percentage of the diet, breeding grain (kernels) to have a higher concentration of critical vitamins is an important method for sustaining a nutritious diet. The International Maize and Wheat Improvement Center (CIMMYT) has developed a panel of maize lines for breeding high grain concentrations of provitamin A carotenoids in varieties that are adapted to tropical and subtropical environments (Suwarno et al. 2015). Genomic prediction (GP) is the technique of building a model that uses genetic markers located throughout the genome as predictors to estimate breeding values for phenotypic traits.
Through predictive modeling, breeders can more efficiently make selections for which lines to advance and/or which crosses to make and, ideally, reduce the time it takes to develop new varieties. This project assesses the accuracy of several genomic prediction strategies within and between four environments in Mexico. Several methods, notably including Ridge Regression and a Reproducing Kernel Hilbert Space (RKHS) approach, have high accuracy in predicting all tested provitamin A carotenoid traits. Furthermore, we show that prediction accuracy is higher using the models that include genome-wide markers than those that only include markers associated with previously-identified carotenoid-related genes, suggesting that GP is worthwhile for these traits in this population. Additionally, we show that a plugin for Genomic Prediction included with a prominent maize genetics (Trait Analysis by aSSociation, Evolution, and Linkage; TASSEL) software performs as well, and in some cases better, than other more computationally intensive methods (Bradbury et al. 2007; Diepenbrock and Bradbury 2015). This plugin is an exciting resource to make the implementation of Genomic Prediction immediately accessible to scientists with fewer computational resources.
Abstract Author(s): Mary-Francis LaPorte, Christine Diepenbrock