Prediction of blown pack in vacuum-packaged beef based on microbiome profiles and supervised machine learning

Area of Science:

• Food microbiology
• Computational biology
• Data science

Background:

• Vacuum-packaged beef spoilage is a challenge, often indicated by the blown pack phenomenon.
• Spoilage is driven by microbial gas production, impacting product shelf life and safety.
• Predictive modeling of spoilage requires understanding microbial dynamics and environmental factors.

Purpose of the Study:

• To investigate the utility of next-generation sequencing (NGS) microbiome analysis and machine learning for predicting blown packs in vacuum-packaged beef.
• To identify key microbial players and environmental factors associated with spoilage and blown pack formation.
• To model the relationship between initial microbiome composition, storage temperature, and spoilage outcomes.

Main Methods:

• Vacuum-packaged beef samples (n=10) were stored at 4°C and 15°C.
• Microbiome profiling was performed using NGS at multiple time points (0, 7, 14, 21, 28 days).
• Machine learning models (XGBoost, Random Forest) were employed for prediction and interpretation using SHAP (Shapley Additive Explanations).

Main Results:

• Blown pack occurrence was predictable using initial microbiome data and storage conditions.
• Temperature was identified as a primary factor influencing blown pack prediction from initial microbiome data.
• Peptoniphilus, Hafnia, and Peptostreptococcus were highlighted as significant bacterial genera associated with spoilage over time.

Conclusions:

• NGS-based microbiome analysis combined with machine learning offers a powerful approach for predicting spoilage in vacuum-packaged beef.
• Understanding the interplay between microbial communities, temperature, and storage duration is crucial for managing blown pack phenomenon.
• These predictive models can be extended to other meat products and storage scenarios to enhance food preservation strategies.