Development of a Rapid Discrimination Tool for Luzhou-flavor Pit Mud Classification by the Kohonen Artificial Neural Network Model

Abstract

Understanding the volatile composition of pit mud (PM) samples and discriminating these samples has become a highly necessary task, owing to the fact that volatile profiling of PM can significantly affect Chinese Luzhou-flavor liquor quality. In this study, the volatile constitutions of 13 Luzhou-flavor liquor PM samples from four typical producing regions were investigated by gas chromatography–mass spectrometry (GC-MS). Owing to their high concentrations, compounds such as ethyl hexanoate, butyric acid, hexanoic acid, ethyl pentadecanoate, ethyl palmitate, ethyl (9E)-9-octadecenoate, ethyl (9E,12E)-9,12-octadecadienoate, and palmitic acid were considered to be predominant volatiles. A promising artificial neural network model, the Kohonen self-organizing map (SOM), was applied to rapidly discriminate the PM samples in terms of differences based on the quantitative information of volatile compounds. After Kohonen SOM training, 13 distinct clusters, corresponding to PM samples, were clearly visualized on a uniform distance matrix (U-matrix). The influence of volatile compounds on the classification of the PM could be displayed using component panels, which can give quantitative insight. GC-MS coupled with the Kohonen SOM model not only presented the volatile constitution of PM but also provided promising information for discrimination between different PM samples, even in other fermented foods.