Data driven techniques to develop new phenotypes for dairy cows and chains

Challenges

Large advances in animal and forage genetics over many decades have driven the profitability of dairy farming, with increased productivity linked to improved cow health and fertility. These advances in dairy productivity have been achieved using data-driven methodologies and technologies. The global climate and biodiversity emergencies recognised by the Scottish Government mean that future developments must maintain profitability while minimising the environmental impacts of dairying, as well as making the sector more resilient to climate change. A key challenge for achieving this is that the traits key for sustainability - feed and lifetime efficiency driven by animal fitness - have been more difficult and costly to measure at a large scale. Addressing the future sustainability of our dairy industries requires us to be able to measure, monitor and manage key performance traits of the animal and the wider system. It is only then that we can identify the underlying factors that impact animal and farm performance, and identify livestock improvement tools.

Machine Learning Methods

Previous studies using milk infrared spectral data showed promising, but not industry-deployable, results in predicting feed intake and methane emission phenotypes. Prediction accuracies were generally low to moderate, especially for predicting methane emissions. Data mining and machine learning methods applied to milk infrared spectral data may provide a more accurate and precise prediction for feed intake. Machine learning methods could improve the quality of the predictions due to their ability to model complex relationships between variables making better use of the increasingly huge data sets and advances in data-driven technology.

Deep learning methods, as a sub-branch of machine learning, have been shown to have good prediction abilities for hard-to-record dairy phenotypes, including feed intake, pregnancy status and bovine tuberculosis status of dairy cows. There is huge potential for applying deep learning methods for phenotype prediction for a wider range of dairy phenotypes, particularly those of higher value as key sustainability indicators. This will allow us to understand the genetics of these traits at scale.

Questions

Solutions

This project aims to use experimental and national farm data to develop new analytical methods to create new predictions, alerts and management tools for dairy cow, herd and supply efficiency, health and sustainability.

Developing and refining the prediction of feed intake and efficiency traits in dairy cows

We are developing new approaches to predict feed intake and other efficiency traits in dairy cows using routinely collected milk spectral data. Building on the long-running Langhill Dairy Experiment, we are using advanced data-driven approaches to explore the best mechanisms to predict these traits within the rich experimental resource. The addition of new records to drive and test these predictions will allow us to work in concert with the industry and iterate to develop and tool that is field-deployable.

Developing and refining the prediction of fitness traits in dairy cows

We are working with the National Milk Records and using their field diagnostic data to extend our understanding of data-driven approaches generated from the experimental herd into commercial settings. Specifically, we are predicting a wider range of fitness traits including Johne’s disease, Bovine viral diarrhoea (BVD) and fertility traits. These predictions allow us to explore the genetics of the traits within the experimental and national herds, and how these traits can be incorporated into genomic breeding value estimation for the national population.

Role of genetic and genomic selection in the improvement of dairy cow efficiency and fitness

The genetic background of animals' feed intake and efficiency is being explored by integrating animals' predicted phenotype for this trait with complex pedigree information and genomic information. Genes and biological pathways that influence animals' feed efficiency are being detected to draw a global picture of genomic architecture controlling this key sustainability trait. This is leading to more accurate genetic prediction for feed efficiency for each animal using accurately predicted phenotype and biological information.

Overall, this project is creating breeding values that are likely to help focus attention on resource use efficiency in general and allow an informed approach to reducing greenhouse gas emissions from dairy production.