Raman Spectra-based Structural Classification Analysis of Flavones, Flavonols, and Isoflavones Using Machine Learning

Abstract

Background: Different C-3 substituted flavonoids have different biological activities and applications in food pharmacology, toxicology, and medicine. Thus, the rapid identification and classification of substitution patterns at C-3 of flavonoids can benefit the processing of flavonoid-related food and medicine. Objective: This study aimed to classify flavonoids with different C3 substituents using Raman spectroscopy, providing a feasible approach for identifying flavonoids in plants. Methods: Eighteen flavonoid samples were selected and dissolved in different solvents. The corresponding Raman spectra were collected by a portable Raman spectrograph. After preprocessing, feature reduction and machine learning were used for the accurate classification of three flavonoids based on 66 Raman spectra. Results: The signals of flavone at 1002, 1245, 1590, and 1609 cm-1 were identified as the characteristic peaks. Peaks at 1298, 1586, and 1605 cm-1 were the special features observed of flavonol. The fingerprint features of isoflavone appeared at 894, 1227, 1321, and 1620 cm-1. All combinations achieved a good classification accuracy of 85%, and the accuracy of the neural network reached 93.3%. Conclusion: The results have demonstrated machine learning to be applicable for the detection and classification of C-3 substituted flavonoids and that feature reduction can aid in the discrimination of Raman spectra variations among diverse C-3 substituted flavonoids, thereby facilitating their further application.