Marker Compound-independent Liquor Identification Enabled By HS-SPME/GC X GC/TOF-MS And Machine Learning Algorithms

The variety of liquors is abundant,and it is one of the most complicated products.Reliable and convenient detection methods are technical premises for guaranteeing drinking safety and quality of Chinese liquors.Traditional detection techniques mainly rely on the analysis of the marker components.On one hand,it can only reflect partial characteristics of a sample.Sensory evaluation based on human organs is time-consuming and labor-intensive.It is an indispensable method in brewing industry.On the other hand,characterized marker components could be easily used by criminals for counterfeit.Therefore,it is necessary to develop a new marker compound-independent liquor identification method for revealing the essential characteristics.Using Chinese liquor as the model for alcoholic beverages in this study,we compared LLE and SPME sample pretreatment methods,carried out P-B test design,and established a standardized detection method.The results showed that SPME offers better sensitivity and robustness.P-B test design indicated that none of the five impact factors affects the extraction efficiency of SPME.Using the standardized method to analyze six different flavor types of liquors,and more than 1,600 components could be identified in each tested liquor.A further PLS analysis was performed,and 338 compounds were found to be closely related to the flavor of Chinese liquors.Two hundred and sixty-two kinds of Chinese liquors were analyzed by comprehensive two-dimensional gas chromatography time-of-flight mass spectrometry.Two-third of the obtained data were randomly picked as training set for machine learning,whereas the remaining 1/3 were treated as test set.The training set was used to optimize the model by machine learning algorithm,and the test set was used for classification analysis with the established model.The results showed that the model accuracy for identification of flavors,brands,origins and vintages is higher than 90%with good reproducibility.This study is a successful attempt to identify alcoholic beverages by integration of machine learning and modern instrumental analysis.It provides the ideas for identification complex composition samples,and it can also provide technical support for the quality control,authentic identification and quality identification of food.