Harnessing Machine Learning and Molecular Docking to Decode the Fatty Acid Dynamics in High-Altitude Yak Milk

This study investigated the fatty acid profile of Muli yak (Bos grunniens) milk and its relationship with compositional parameters across different parities. Milk samples from second-, third-, and fourth-parity yaks were analysed for protein, fat, vitamins, minerals, and 37 fatty acids using gas chromatography. Statistical analyses included ANOVA, correlation analysis, principal component analysis (PCA), machine learning algorithms, and molecular docking. Parity significantly affected 15 components (p < 0.05), with third-parity milk showing the highest eicosapentaenoic acid (EPA, C20:5n3) and arachidonic acid (ARA, C20:4n6) concentrations. Among 134 significant correlations, calcium-ARA and ARA-EPA exhibited strong positive associations (|r| > 0.67). PCA explained 54.2% of the variance through three principal components, differentiating samples by parity. The optimal prediction models were ARA-XGBoost, EPA-Random Forest, ALA-GAM, and LA-SVM, with calcium and protein serving as key predictors. Molecular docking revealed that EPA-FABP2 had the lowest binding energy. These parity-related shifts in functional long-chain polyunsaturated fatty acids are meaningful for the nutritional value of yak milk (e.g., omega-3/omega-6 profile) and may also influence technological properties associated with milk fat composition (e.g., oxidative stability and processing behaviour), supporting parity-oriented quality evaluation and targeted utilisation of yak milk.