Analysis and content prediction of sugar metabolites in G. elata using a combined approach of GC-MS, FTIR, NIR and machine learning

The quality of Gastrodia elata Blume ( G. elata ), used for both food and medicine, closely relates to the sugar metabolites present in the plant. This study employed a combination of analytical techniques, including gas chromatography-mass spectrometry (GC-MS), Fourier transform infrared spectroscopy (FTIR), and near-infrared spectroscopy (NIR), alongside machine learning methods, to systematically analyze and predict the content of sugar metabolites in G. elata samples from various sources. The results show that the sugars in G. elata are mainly monosaccharides, with sucrose, glucose, and D-fructose being the predominant components. While the sugar metabolites of samples from different geographical regions are similar, significant differences exist in their content. The Long Short-Term Memory (LSTM) model built using FTIR and NIR and their data fusion (FTIR + NIR) demonstrated acceptable quantitative prediction performance for 25 sugar compounds; however, it failed to provide reliable predictions for three compounds (L-fucose, L-rhamnose, and 2-deoxy-D-ribose). Within the scope of the existing sample set, this study demonstrates the preliminary feasibility of using LSTM models for the quantitative prediction of sugar metabolites in G. elata . Future research can build on this foundation to optimize the model further and expand its scope of application.