Volatilomic Profiling of Juices by Dual-Detection HS-GC-MS-IMS and Machine Learning-An Alternative Authentication Approach.

A prototype dual-detection headspace-gas chromatography-mass spectrometry-ion mobility spectrometry (HS-GC-MS-IMS) system was used for the analysis of the volatile profile of 47 juices including grapefruit, blood orange, and common sweet orange juices without requiring any sample pretreatment. Next to reduced measurement times, substance identification could be improved substantially in case of co-elution by considering the characteristic drift times and / ratios obtained by IMS and MS. To discriminate the volatile profiles of the different juice types, extensive data analysis was performed with both datasets, respectively. By principal component analysis (PCA), a distinct separation between grapefruit and orange juices was observed. While in the IMS data grapefruit juices not from fruit juice concentrate could be separated from grapefruit juices reconstituted from fruit juice concentrate, in the MS data, the blood orange juices could be differentiated from the orange juices. This observation leads to the assumption that the IMS and MS data contain different information about the composition of the volatile profile. Subsequently, linear discriminant analysis (LDA), support vector machines (SVM), and the -nearest-neighbor (NN) algorithm were applied to the PCA data as supervised classification methods. Best results were obtained by LDA after repeated cross-validation for both datasets, with an overall classification and prediction ability of 96.9 and 91.5% for the IMS data and 94.5 and 87.9% for the MS data, respectively, which confirms the results obtained by PCA. Additional data fusion could not generally improve the model prediction ability compared to the single data, but rather for certain juice classes. Consequently, depending on the juice class, the most suitable dataset should be considered for the prediction of the class membership. This volatilomic approach based on the dual detection by HS-GC-MS-IMS and machine learning tools represent a simple and promising alternative for future authenticity control of juices.