Machine learning-based discovery of informative SNPs for population assignment through whole genome sequencing

Area of Science:

• Animal Genetics
• Bioinformatics
• Machine Learning

Background:

• Whole-genome sequencing (WGS) provides extensive genetic data for farm animals.
• Machine learning (ML) algorithms can identify population-informative genetic markers.
• Accurate population assignment is crucial for genetic resource conservation and breeding optimization.

Purpose of the Study:

• To evaluate ML-based methods for identifying informative single nucleotide polymorphisms (SNPs) for water buffalo population assignment.
• To compare the performance of different SNP selection algorithms and ML classifiers.
• To determine the optimal SNP set size for accurate breed classification.

Main Methods:

• Analyzed WGS data from 187 water buffaloes across eight breeds.
• Employed SNP selection methods: F_ST, Minimum Redundancy Maximum Relevance (mRMR), and Relief-F.
• Utilized ML classifiers: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Naive Bayes (NB), and Random Forest (RF).
• Assessed accuracy across various SNP set sizes (100 to 10,000).

Main Results:

• Hierarchical population structure was identified, separating river and swamp type buffaloes.
• SVM and NB classifiers showed superior assignment accuracy compared to RF and KNN.
• Accuracy increased with SNP number, exceeding 85% with 10,000 SNPs.
• mRMR with SVM achieved 97.6% accuracy using 2000 SNPs; 768 mRMR SNPs yielded 98.8% accuracy.

Conclusions:

• An effective framework integrating ML feature selection and classification was developed for genetic marker mining.
• This approach enables reliable population assignment using WGS data.
• The study highlights the utility of mRMR and SVM for water buffalo breed identification.