New strategies for the differentiation of fresh and frozen/thawed fish: A rapid and accurate non-targeted method by ambient mass spectrometry and data fusion (part A)

Abstract

Ambient mass spectrometry (AMS) was applied for the first time to differentiate fresh and frozen/thawed Dicentrarchus labrax (European sea bass). One hundred and twenty samples were submitted to two extraction procedures and analyzed in positive and negative ion modes by direct analysis in real time-high resolution mass spectrometry (DART-HRMS). The four DART-HRMS datasets were concatenated with a low-level data fusion approach, and the resultant unique block of data was submitted to multivariate statistical analysis to tease out the most informative m/z values capable of codifying fresh and defrosted D. labrax . The statistical significance (p-value) and fold change (FC) of these informative signals were then evaluated with univariate analyses and tentatively assigned. The 25 features were then used to build a support vector machine (SVM) model capable of classifying D. labrax samples according to whether they were fresh or previously frozen. The SVM model has accuracy, sensitivity and specificity of 100%, both in training and test set. The SVM model was used to successfully classify an independent set of twenty fresh and frozen/thawed Salmo salar (Atlantic salmon). The concentrations of the most significant molecular features were quantified by gas-chromatography-mass spectrometry. • (+/−)DART-HRMS of polar and non-polar extracts of sea bass was performed. • The four DART-HRMS datasets were concatenated by low-level data fusion. • PLS-DA was performed on the training set to extrapolate the informative signals. • SVM classification model was constructed with the selected signals. • The built-in SVM model was validated with on hold-out subset with high accuracy.