(Internal) Evaluating the importance of relatedness in genomic prediction

Advisors: Lewis Lukens, Plant Agriculture

Proposed co-advisor (UoG): TBC

The key to increasing food production for a growing population has been genetic improvement. Genetic improvement in many crops involves crossing two elite cultivars, evaluating 100s or 1000s of their progeny, and selecting those with the desired traits. Increasingly, researchers sequence progeny SNPs and provide these to genomic prediction models. These models’ DNA-based predictions of progeny traits are then used for selection. A difficulty is that a genomic prediction model’s prediction quality depends on the relatedness of genotypes used to train the model and the genotypes that are to be predicted. Models trained with genotypes closely related to selection targets tend to predict target traits well. Models trained on genotypes distantly related to the selection targets predict target traits poorly. The objective of this work is to quantify the effect of relatedness on genomic prediction quality. We have SNP genotyped and evaluated large full-sib, half-sib, and unrelated barley populations. Statistical and machine learning models will evaluate how training and testing-population relatedness affects genomic predictions. This work will help generate new barley varieties and contribute to quantitative genetics research.

This is a one-semester project.