Smart animal breeding with advanced machine learning
The amount of data that is accumulating over the lifespan of farm animals is increasing very rapidly, but the technology to analyse it to full potential is lagging behind. We propose to investigate the applicability of advanced machine learning methods to get insight into the variations in phenotypic patterns over time. In particular, we will investigate if adding phenotypic data over the life history of an animal to the genomic information will improve the prediction of certain phenotypes. We will use predictive methods, such as the Generalized Matrix Learning Vector Quantization (GMLVQ) and decision trees, to predict probabilities that the performance of an animal will fall into a predefined set of classes. We will also investigate ensemble methods that combine the results of different predictive models.
The main contributions of this project are two-fold; to improve the well-being of farm animals and to improve their performance. On the technical we will develop:
- A method that identifies animals with health status that (temporarily) deviates from what is expected. This would allow for early prediction and identification of reduced health so that timely action can be taken. It would also allow to selectively breed for increased health status at no additional costs.
- A method to that predicts the output performance of an animal. This would allow for pre-sorting of animals, for example finisher pigs to groups targeted at intended markets, and determining cows that will reach second lactation.
Breed4Food STW Partnership
The Breed4Food STW Partnership “Predicting Phenotypes” research projects are:
- From sequence to phenotype: detecting deleterious variation by prediction of functionality
- GenoMIX: utilizing crossbred information to accelarete genetic progress
- SmartBreed: Smart animal breeding with advanced machine learning
- Topbreed; towards precision breeding using genomic prediction
Mutual goal is a more sustainable and animal-friendly breeding.