Optimal Use Of Phenotypic Data For Breeding Using Genomic Predictions

Heslot, Nicolas Didier

Optimal Use Of Phenotypic Data For Breeding Using Genomic Predictions

Abstract

Genomic predictions or genomic selection (GS) was proposed to overcome a number of challenges in application of marker assisted selection to complex quantitative traits. Simulations and empirical studies suggest that GS can improve genetic gain per unit time and cost. The cost of molecular markers has dramatically decreased over the past 10 years and should continue to do so with progress in sequencing technologies whereas phenotyping cost should remain stable or increase with land and labor costs. This means that the most valuable and limiting part in breeding will increasingly be the phenotype and not the genomic data. As a consequence, it is critical to make the most of the scarce phenotypic data available. GS opens numerous possibilities to do so. First, using eight wheat (Triticum aestivum L.), barley (Hordeum vulgare L.), Arabidopsis thaliana (L.) Heynh., and maize (Zea mays L.) datasets, the predictive ability of currently available GS models was evaluated by comparing accuracies, the genomic estimated breeding values (GEBVs), and the marker effects for each model. While a similar level of accuracy was observed for many models, the computation time varied widely as did the distribution of marker effect estimates. Second, allele replication rather than genotype replication was investigated as a new way to cope with highly unbalanced phenotypic data sets. Using a two-row elite barley (Hordeum vulgare L.) population from a commercial breeding program, I demonstrated the possibilities offered by GS to analyze multienvironment trials, identify outliers, group environments, and select historical data relevant for current breeding efforts. Finally, we proposed, developed and tested a new model to use environment data to model genotype by environment interactions (G*E) in GS. A crop model was used to derive stress covariates from daily weather data for predicted crop development stages. I extended the factorial regression model to genomic selection. Machine learning was also used to capture non-linear responses of QTL to stresses. The method was tested using a large winter wheat dataset. This new model provides insight into the genetic architecture of genotype by environment interactions and could predict genotype performance based on past and future weather scenarios.

Date Issued

2014-01-27

Keywords

genomic selection; genotype by environment interactions; genomic predictions

Committee Chair

Sorrells, Mark Earl

Committee Co-Chair

Jannink, Jean-Luc

Committee Member

Mezey, Jason G.

Degree Discipline

Plant Breeding

Degree Name

Ph. D., Plant Breeding

Degree Level

Doctor of Philosophy

Types

dissertation or thesis