Interpretive Summary: Categorization of birth weight phenotypes for inclusion in genetic evaluations using a deep neural network
By Anne Zinn
In the United States, phenotypic data is used for genetic evaluations of beef cattle. As the list of genetic merit predictions increase, so does the labor required to collect and record various phenotypes. For example, birth weight serves as a valuable indicator of the economically relevant trait of calving ease and erroneous data collection for birth weight could impact genetic evaluations for calving ease, but the collection of this data could cause economic strain. A study recently published in the Journal of Animal Science characterized birth weight phenotypes for inclusion in genetic evaluations using a deep neural network. The objective of the current study was to evaluate the use of deep neural networks to categorize contemporary groups based on data quality and to determine the impact of removing data predicted as fabricated on the ranking of animals for calving ease expected progeny differences.
Results showed that the prediction accuracy of the deep neural network models trained to classify contemporary groups based on the data generation process for birth weight was dependent on both the number of neurons in the hidden layers and on the activation functions used in these neurons. In terms of accuracy and consistency, the best performing deep neural network model used an increased number of neurons in the hidden levels and used the sigmoid activation function. Additionally, the classification of the contemporary groups from the Hereford population was consistent.
In conclusion, the use of deep neural networks as an objective method of classifying data quality prior to the inclusion of genetic evaluations represents a tangible use of artificial intelligence in routine genetic prediction. Moving forward, additional research is justified and warranted for other traits.
The full paper can be found on the Journal of Animal Science webpage.