Location: Plant, Soil and Nutrition Research
Project Number: 8062-21000-038-00
Start Date: Apr 01, 2013
End Date: Mar 31, 2018
Develop Bayesian analogues of the factor analytic models and extended for genomic prediction. The Bayesian formulations will enable model averaging thereby minimizing required user input. Analysis output will be processed to increase its interpretability. Modify these models to fit random marker effects, including for cases when many marker effects are missing. Test hypotheses relative to the value of the generalized coefficient of determination (GCD) of the mean contrast between parents of the selection candidates and the training population mean for genomic prediction accuracy for those candidates. Develop an algorithm to estimate the GCD of the expected mean contrast between selection candidates themselves and the training population mean using the distribution of possible candidate genotypes constructed from knowledge of their parental genotypes. Extend the GCD approach to cases where marker effects are not all assumed to come from the same normal distribution. Develop new breeding schemes that leverage genomic data to optimally balance short- and long-term genetic gain. Develop optimization algorithms that use genetic variance components, economic costs and budgets, and logistic constraints to compute a plan specifying the number of crosses to be made, the number of lines to develop per cross, and the fraction of lines to genotype to maximize gain from selection in a multitrait or polygenic trait context. Test hypotheses relative to the value of whole-population marker imputation accuracy as a predictor of whether multi- or specific-population training should be used to predict genotypic or breeding values of progeny admixed by introgression of exotic germplasm. Test methods to retain diversity at the breeding program level during genomic selection, one that works to minimize relatedness in the selected set and one that incorporates an estimate of genetic potential of the selected set. In conjunction with these methods, estimate chromosome segment-specific levels of repulsion phase linkage disequilibrium to select individuals carrying effective recombination events. Assemble these methods of genomic prediction into a single package in the free, open-source, statistical language R to facilitate user access to the methods. Integrate methods into a database constructed specifically to house breeding line performance and genotype data. Develop download methods and visualizations for analysis results.