AUTHOR=Mai Ruijie , Xue Shuo , Ren Jingnan , Fan Gang , Yang Jinchu , Huang Linhua , Li Guijie , Cheng Yujiao , Wang Qiuling , Yang Yongfeng , Huang Zhenzhen , Feng Yingjie , Liu Wenzhao , Yu Aiqun , Feng Jian 

TITLE=Mechanism and Multilayer Perceptron prediction model of the removal of α-terpineol, terpinen-4-ol and carvone from pasteurized citrus juices by β-cyclodextrin encapsulation

JOURNAL=Frontiers in Nutrition

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/nutrition/articles/10.3389/fnut.2025.1557934

DOI=10.3389/fnut.2025.1557934

ISSN=2296-861X

ABSTRACT=This study revealed the mechanism of improvement in sensory evaluation of citrus juices after pasteurization by the natural cyclodextrin (β-CD), and developed a prediction model for encapsulation based on conventional physicochemical indicators. The results of gas chromatography–mass spectrometry indicated that the off-flavor in citrus juice after pasteurization were mainly caused by α-terpineol, terpinen-4-ol and carvone, and the addition of β-CD could effectively reduce the content of these compounds. The inclusion complexes of β-CD and off-flavor compounds were characterized by formation constants, scanning electron microscope, X-ray diffraction, Fourier transform infrared spectroscopy, and thermogravimetric analyses, followed by molecular docking to show the possible conformations of β-CD and off-flavor compounds to form 1:1 inclusion complex by hydrophobic interaction, van der Waals forces and hydrogen bonding. Multiple prediction models were constructed and evaluated by deep learning using basic physicochemical indicators such as sucrose content, citric acid content, pH, temperature and storage conditions as input variables, and peak area of off-flavor compounds as output layer. The results showed that the Multilayer Perceptron Model had great potential in predicting the cyclodextrin embedding effect.