Machine learning to further improve the decision which boar ejaculates to process into artificial insemination doses

Area of Science:

• Animal Science
• Reproductive Biology
• Data Science in Agriculture

Background:

• Current artificial insemination (AI) practices rely on limited computer assisted semen analysis (CASA) parameters for boar ejaculate selection, leading to variable fertility outcomes.
• Existing AI protocols may not fully capture the complex factors influencing boar fertility, necessitating improved predictive models.

Purpose of the Study:

• To investigate if incorporating additional data sources with machine learning models can improve the prediction of boar ejaculate fertility.
• To identify novel parameters beyond standard CASA metrics that influence fertility outcomes in AI doses.

Main Methods:

• Gradient Boosting Machine (GBM) models were developed to predict eight fertility phenotypes (gestation length, total born, born alive, stillborn) using data from CASA, boar ejaculate characteristics, breeding values, and weather.
• Data were divided into training, validation, and independent test sets for robust model evaluation.
• Seven prediction sets were created, incrementally adding data sources to assess their predictive power.

Main Results:

• The GBM models, even with expanded datasets, showed low predictive performance for fertility phenotypes, failing to improve upon current AI dose quality checks.
• The hypothesis that additional data and GBM models would enhance AI dose selection was rejected.
• Several parameters not currently used in routine AI processing, including CASA, breeding value, and weather-related factors, were identified as potentially significant predictors of boar fertility.

Conclusions:

• The study did not validate the hypothesis that increased data and GBM models improve AI dose quality assessment.
• Current AI laboratory practices for boar ejaculate selection may not be significantly enhanced by the tested data-driven approach.
• Identified novel parameters warrant further investigation for their potential to refine boar ejaculate quality assessment in AI.