Comparing assignment-based approaches to breed identification within a large set of horses

Area of Science:

• Animal genetics and bioinformatics
• Application of machine learning in animal breeding

Background:

• Animal breeding increasingly utilizes extensive data sets and statistical techniques, integrating machine learning.
• Assessing genetic diversity and population structure is crucial for effective breeding programs.

Purpose of the Study:

• To evaluate the potential of machine learning and data mining for horse breed identification.
• To compare various individual assignment methods and their success factors in a Czech horse population.
• To analyze genetic differentiation and gene flow among horse breeds.

Main Methods:

• Utilized 314,874 allelic data sets from various horse breeds.
• Compared eight standard methods from GeneClass software (Bayesian, frequency, distance).
• Evaluated mainstream classification algorithms from the WEKA machine learning workbench (e.g., Bayes Net, Random Forest, SVM).

Main Results:

• The Bayesian method in GeneClass (89.9%) and the Bayesian network algorithm in WEKA (84.8%) showed superior performance in individual assignment.
• Genetic differentiation varied among breeds, with fixation index values ranging from 0.057 to 0.144.
• Highest genetic divergence was observed between Friesian and Equus przewalskii; highest gene migration was between Czech and Bavarian Warmblood.

Conclusions:

• Machine learning and data mining tools, particularly Bayesian approaches, are effective for assessing breed traceability and aiding breeding management decisions.
• Breed genomic prediction accuracy was highest in cold-blooded horses.
• The accuracy of individual assignment is influenced by the number of breeds and their genetic divergence.