Integrative lipidomics profiles and explainable machine learning for species-specific identification of meat and bone meal.

Accurate species identification of meat and bone meal (MBM) is a regulatory requirement in many jurisdictions to prevent intra-species recycling (cannibalism) in animal feed. It is a practice that facilitates transmission of bovine spongiform encephalopathy via prion-contaminated feed. This study represents the first comprehensive application of nontarget lipidomics method to systematically characterize the lipid composition profiles across 76 MBM samples from four species. A total of 4180 lipid molecules were identified, classified into 43 lipid subclasses. Through rigorous evaluation criteria (P < 0.05, VIP>1.00, FC>1.50 or FC<0.67), 46 highly species-specific lipid biomarkers were successfully screened. Leveraging the abundance profiles of these discriminative lipids, we established optimized species identification models using multiple machine learning algorithms. The support vector machine model demonstrated superior performance with 100% classification accuracy. Subsequent SHapley Additive exPlanations analysis revealed that three lipid molecules, TG(44:0), TG(49:0), and Cer(t41:1), contributed most significantly to species differentiation. The developed analytical framework, combining lipidomics profiles with explainable machine learning, establishes a scientific foundation for precise species identification in MBM.