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

The preservation of vacuum-packaged beef products is essential for maintaining shelf life. However, the occurrence of blown pack phenomenon, characterized by the expansion of packaging due to gas production by spoilage microorganisms, is still a challenge. In the present work, we demonstrate that microbiome analysis using next generation sequencing (NGS) and machine learning might be useful in the analysis, modeling and prediction of spoilage and blown pack in vacuum-packaged beef. Beef systems (n = 10) were vacuum-packed, stored at 4 and 15 °C, and their populations were monitored based on NGS at 0 h and 7, 14, 21 and 28 days. Our analysis allowed the prediction of blown pack based on information of the initial microbiome in beef and storage conditions, identification of the relationship of different bacteria genera associated with spoilage along with temperature, which were consistent with differential abundance analysis, and estimate the relationship of temperature and blown pack. Using SHAP (Shapley Additive Explanations) to interpret the XGBoost model, we identified temperature as the most influential factor in blown pack prediction when considering microbiome data from day zero. Additionally, SHAP analysis of Random Forest and XGBoost models based on OTU Spearman correlation and linear regression, computed about time, highlighted Peptoniphilus as the most important bacterial genus, followed by Hafnia and Peptostreptococcus. Additional studies might extend these methods for other types of meat, cuts and including additional storage conditions, allowing a better modeling of the dynamics in the microbiome associated with the blown pack phenomenon.