Data integration strategies for genomic prediction and association studies in wheat breeding

Abstract

Advancements in genotyping and phenotyping technologies have allowed for progress in data-driven wheat breeding. Resulting phenotypic and genotypic data can improve the prediction of promising variety candidates if a unified evaluation across data silos succeeds. This thesis explores integrative strategies for genomic prediction in wheat hybrids and inbred lines, as well as for genome-wide association mapping. Common checks and methodology in existing wheat breeding programs proved beneficial for integration. Genome-wide association studies were found to yield fewer significant marker-trait associations than for individual data sets, albeit with higher predictive power. The predictive power of genomic prediction increased markedly, showing decreasing additional benefits as dataset sizes grew. This shows that combining data across silos is beneficial, but unresolved factors remain that limit the predictive power. This is potentially due to genotype-times-environment interactions, which are difficult to track due to the strongly imbalanced data. Methodological innovations, like balanced environmental sampling, can be further explored based on initial results from this work.