C/H/O stable isotope of rape honey and its components combined with machine learning algorithms to identify geographical origin

RationaleThe distribution of rape honey is among the largest and most diverse of all honeys available to humankind with respect to the geographical origin. Accurate isotopic reference values for rape honey are therefore important for precise verification of honey origin and its traceability. New combined rape honey delta C-13, delta H-2, and delta O-18 values in combination with values on its compounds (protein and saccharides) were used to complement existing databases to better identify the geographical origin of Chinese rape honey. MethodsTraceability methods based on elemental analyzer isotope ratio mass spectrometry and liquid chromatography isotope ratio mass spectrometry were established for geographical origin of rape honey. ResultsRape honey harvested in the high-altitude region (QH; Qinghai) had significantly higher values (1.4 to 5.3 parts per thousand for delta C-13, 7.9 to 12.9 parts per thousand for delta H-2(protein)) for the delta C-13 of whole honey (-23.8 parts per thousand), its protein (-24.4 parts per thousand), fructose (-23.5 parts per thousand), glucose (-23.6 parts per thousand), and disaccharide (-24.7 parts per thousand), and also delta H-2 of the protein (103.5 parts per thousand) than those in low-altitude regions (HB; Hubei, SC; Sicuan, and JS; Jingsu). The delta O-18(rape honey) was a useful index to differentiate whether rape honey from coastal (JS) or non-coastal (HB, SC, and QH) regions. The delta C-13, delta H-2, and delta O-18 values in rape honey are affected by geographical factors, such as temperature and altitude. The delta C-13(protein) and delta C-13(rape honey) values were better to identify the geographical origin of rape honey than delta C-13(saccharides). The delta O-18 and delta H-2 values of rape honey protein were more suitable for traceability than those of rape honey. The combination of the delta C-13, delta H-2, and delta O-18 values of rape honey and its extracted protein and saccharides improved the precision of three models (linear discriminant analysis, SVM, and random forest) used to discriminate rape honey from different regions in China. The SVM model obtained the best accuracy (93.2%). ConclusionsStable isotopes could be significant predictors in determining the geographical origin of rape honey.