Application of machine learning for optimizing biomarker combinations and guiding decisions on meat authentication

Area of Science:

• Agricultural Science
• Food Science
• Computational Biology

Background:

• Meat authentication is crucial for verifying production systems and ensuring product quality.
• Machine learning approaches offer novel methods for analyzing complex biological data in food authentication.

Purpose of the Study:

• To evaluate the efficacy of decision trees (DTs) and random forest models (RFs) for authenticating lamb meat based on production systems.
• To identify key biomarkers and optimize classification strategies for meat authentication.

Main Methods:

• Application of DTs and RFs to analyze 19 variables measured on perirenal fat (PF), dorsal fat (DF), and longissimus thoracis et lumborum (LTL) muscle.
• Experiment involved Romane male lambs finished on pasture or in stalls, with varying durations.

Main Results:

• DTs/RFs achieved up to 95.7% accuracy in distinguishing pasture-finished from stall-fed lamb carcasses.
• PF skatole and PF carotenoid pigment content were identified as prominent classification biomarkers.
• A point-of-sale model using DF spectrocolorimetric and LTL muscle color achieved 85.4% accuracy.

Conclusions:

• DTs and RFs demonstrate significant potential for meat authentication, offering robust classification capabilities.
• The study highlights the importance of specific fat and muscle tissue biomarkers for accurate authentication.
• Further validation on larger datasets is recommended for broader application of DTs in meat authentication.