Could machine learning predict meat quality traits using the metabolic fingerprints of lamb meat analysed with REIMS?

Abstract

High pH is a quality defect in meat and is often a result of pre-slaughter stress. We have developed a farmyard stress model that is a useful tool for research on high ultimate pH meat [1]. In this study, we used Rapid Evaporative Ionisation Mass Spectrometry (REIMS) to fingerprint the metabolites within the meat [2]. REIMS data was able to distinguish between muscles of lambs that have been exercised pre-slaughter compared to those who have not [2]. Most REIMS data analysis is done using multivariate statistics. These methods are powerful but lack inbuilt methods to refine the REIMS dataset to focus on only those variables that are important for explaining the phenomenon of interest. This is critical for methods such as REIMS where most of the data collected is irrelevant to the study question and interferes with model validity. Iterative PLS-DA and machine learning have been used for REIMS analysis but their application has been limited to date. Machine learning (ML) is a novel and emerging tool incorporating artificial intelligence to train the model to predict the output with algorithm input. The potential of advanced machine learning compared to conventional multivariate statistical analysis has been shown in classifying beef cuts from images, fish speciation based on REIMS features and determining chicken freshness via colorimetric sensor array. The current study evaluated using WEKA and Tensorflow platforms to predict ultimate pH and meat quality parameters from the REIMS metabolic fingerprint. These data will contribute towards defining and designing much larger experiments run in commercial processing plants to develop rapid screening systems to predict future pH as meat is processed.