Classification accuracy of machine learning algorithms for Chinese local cattle breeds using genomic markers

Area of Science:

• Animal Genetics
• Bioinformatics
• Machine Learning

Background:

• Accurate farm animal breed classification is crucial for genetic resource conservation.
• Phenotypic methods struggle with distinguishing similar breeds.
• Genomic data offers a powerful alternative for breed classification.

Purpose of the Study:

• To evaluate machine learning algorithms for classifying Chinese local cattle breeds using genomic SNP data.
• To compare the effectiveness of different feature selection methods (F_ST, mRMR, Relief-F) and machine learning algorithms (SVM, Random Forest, Naive Bayes).

Main Methods:

• Utilized genomic SNP data from 213 individuals across seven Chinese local cattle breeds.
• Compared three feature selection methods: F_ST value sorting, mRMR, and Relief-F.
• Evaluated three machine learning algorithms: Support Vector Machine (SVM), Random Forest, and Naive Bayes.

Main Results:

• SVM achieved >99.47% accuracy using F_ST (>1500 SNPs) or mRMR (>1000 SNPs) for feature selection.
• SVM was the most effective algorithm, followed by Naive Bayes (NB).
• F_ST and mRMR were the best SNP selection methods, outperforming Relief-F. Misclassification occurred between highly similar breeds.

Conclusions:

• Machine learning models combined with genomic data provide effective, rapid, and accurate classification of local cattle breeds.
• This approach offers a technical foundation for cattle breed classification in China.
• Highlights the potential for genomic-based methods in farm animal genetic resource management.