Abstract
Selecting an animal for breeding requires knowledge of the utility it passes on to its descendants, known as a breeding value. Improvements in technology mean that genetic material can be directly measured, creating what is known as a genotype. The use of these data has improved the accuracy of breeding value prediction, but many of datasets of interest contain a mixture of genotyped and ungenotyped animals. It is challenging to make good use of all available data to best predict breeding values while allowing for missingness. We consider some limitations of present approaches, and propose several approaches intended to improve on these weaknesses.