Using phenotypic distribution models to predict livestock performance

Area of Science:

• Animal Science
• Agricultural Technology
• Machine Learning in Agriculture

Background:

• Indigenous livestock breeds are locally adapted but may have lower productivity.
• Commercial breeds often underperform in new environments due to genotype by environment interactions.
• Predicting commercial breed performance in diverse regions like sub-Saharan Africa is challenging.

Purpose of the Study:

• To develop a novel methodology for modeling livestock performance using growth data.
• To predict the suitability of commercial chicken breeds in various Ethiopian agro-ecologies.
• To identify key environmental variables influencing breed productivity.

Main Methods:

• Utilized growth data from chicken breeds tested in Ethiopia.
• Employed machine learning algorithms to build phenotype distribution models.
• Predicted body weight as a function of environmental variables to assess breed suitability.

Main Results:

• Successfully predicted chicken breed performance across different Ethiopian environmental conditions.
• Identified specific environmental factors driving body weight variation for each breed.
• Assigned breeds to optimal agro-ecologies based on predicted body weight.

Conclusions:

• Phenotype distribution models are crucial for predicting livestock productivity in varied environments.
• Acknowledging genotype by environment interactions is vital for livestock breeding strategies.
• This approach can guide the development of breeds better suited to specific production systems.