Metabolomics for origin traceability of lamb: An ensemble learning approach based on random forest recursive feature...

Area of Science:

• Food Science
• Analytical Chemistry
• Biotechnology

Background:

• Consumer demand for traceable lamb products is high.
• Untargeted metabolomics is used for meat origin traceability, but biomarker identification is challenging.
• Accurate lamb breed traceability is crucial for quality assurance and consumer trust.

Purpose of the Study:

• To develop a rapid and accurate method for identifying metabolic biomarkers for lamb origin traceability.
• To evaluate the effectiveness of machine learning algorithms in analyzing metabolomics data for breed-specific traceability.
• To establish a robust panel of biomarkers for geographical indication lamb traceability.

Main Methods:

• Untargeted metabolomics was applied to analyze five breeds of geographical indication lamb.
• Random forest recursive feature elimination was used to identify potential metabolic biomarkers from 4139 metabolites.
• A panel of 14 metabolic biomarkers was refined and validated.
• Naive Bayes algorithm was employed to assess classification accuracy.

Main Results:

• 29 potential biomarkers with breed-specific and environment-related variations were initially identified.
• A refined panel of 14 metabolic biomarkers demonstrated high accuracy and robustness for lamb origin tracing.
• The combination of the 14 biomarkers and the Naive Bayes algorithm achieved the highest classification accuracy.
• Random forest recursive feature elimination proved effective for high-dimensional metabolomics data.

Conclusions:

• A panel of 14 metabolic biomarkers, identified using random forest, significantly enhances lamb breed-specific traceability.
• Machine learning-based biomarker panels offer a powerful approach for accurate meat origin traceability.
• This study provides a practical framework for developing advanced traceability systems in the food industry.