Machine Learning Combining with Ultra-High Performance Liquid Chromatography-Time-of-Flight Mass Spectrometry and Gas...

The objective was to achieve accurate origins discriminations of flos of Lonicerae japonicae, especially those from the genuine and non-genuine producing areas. An integration strategy was proposed based on machine learning, combining ultra-high performance liquid chromatography-time-of-flight mass spectrometry and gas chromatography-ion mobility chromatography data fusion for geographical origins discrimination of flos of Lonicerae japonicae. Sixty-one batches of flos of Lonicerae japonicae samples from different origins were determined by ultra-high performance liquid chromatography-time-of-flight mass spectrometry and gas chromatography-ion mobility chromatography. Multivariate statistical analysis, including principal component analysis and partial least-squares discriminant analysis were performed for the classification with an accuracy of 80.33% and 96.72%, respectively. Variable importance in projection (VIP>1) screened 9 nonvolatile differential constituents and 32 volatile differential constituents, respectively. The results of machine learning models combined with ultra-high performance liquid chromatography-time-of-flight mass spectrometry and gas chromatography-ion mobility chromatography data fusion indicated that the multilayer perceptron, logistic, gradient boosting decision tree, and Decision Tree (CART)) combined with a middle-level data fusion approach, achieve 100% classification accuracy with the training set. Among them, the multilayer perceptron algorithm achieved 100% accuracy for both the training and testing sets. These findings provide methodology support for tracing the origin of flos of Lonicerae japonicae.