Research
Our current research focuses on four areas: predictive breeding, heterotic groups in maize, AI via machine learning in selection, and sweet corn.
Predictive breeding
Predictive breeding via genomewide selection (or genomic selection) exploits the availability of cheap and abundant markers. Phenotypic and single nucleotide polymorphism (SNP) marker data in a training population are used to develop a statistical model to predict performance for traits such as grain yield, moisture, and lodging. The accuracy of the genomewide predictions is assessed from a validation population, which has been phenotyped and genotyped.
If the predictions are sufficiently accurate, the statistical model is used to predict the performance of candidates in a test population, which has been genotyped but not yet phenotyped.
Our research on genomewide selection since 1994 has shown that the procedure works and that are is no reason to not use genomewide selection in maize breeding. While we still do some research in genomewide selection, we are moving away from this area of research.
Heterotic groups in maize
Pedigree and DNA analyses have shown that today’s corn hybrids largely trace their origins to only seven lineages. Three lineages form the pool of female parents used in hybrid seed production, and four lineages form the pool of male parents.
This project aims to create new variation among corn hybrids by restructuring the female and male parental pools. We will utilize the newest corn lines whose Plant Variety Protection certificates have expired. With phenotypic and molecular marker data, we will first break the current population structures of U.S. Corn Belt germplasm, then develop new ones.
AI via machine learning in selection
We have found that AI via machine learning can effectively replicate a maize breeder’s selection decisions. With machine learning, the probability that a random selected (i.e., by a breeder) candidate was ranked ahead of a random non-selected candidate was around 90%. This high probability, which did not depend on the proportion selected by a breeder, was supported by around 70-80% of a breeder’s decisions being exactly replicated by machine learning.
These results indicate that a decision-support system for breeders is feasible. We are conducting further expand the use of AI in selection decisions. For example, apple breeders subjectively evaluate individual trees by visually evaluating the candidates and tasting their fruit. Can AI coupled with molecular markers effectively predict the best trees when the selection criteria are subjective?
Sweet corn
Sweet corn breeding in our program focuses on developing supersweet (shrunken-2) cultivars that proudly feature the maroon & gold school colors of the University of Minnesota. Aka GopherCorn.
