Machine Learning for Precise Identification of Royal Jelly From Various Food Sources With Stable Isotope and Physicochemical Properties

Abstract

Royal jelly (RJ) is highly regarded for its bioactive compounds and salutary effects. However, the traceability and authenticity of royal jelly are significantly challenged due to the considerable variability in its composition, which is influencedby the diverse food sources of bees. This study examines the impact of three food sources—natural foods, sugar-water, and pollen substitutes—on stable carbon (δ¹³C) and nitrogen (δ¹⁵N) isotope fractionation in RJ produced during different floral periods. The findings indicate that RJ derived from natural honey and beebread exhibited lower δ¹³C values, whereas RJ produced from sugar-water feeding showed higher δ¹³C values. Furthermore, a notable degree of variation in δ¹³C was observed regarding the diverse beebread sources. A positive correlation was identified between δ¹⁵N in beebreads and RJ, whereas a negative correlation (r = −0.89) was observed between δ¹⁵N in pollen substitutes and RJ. The application of machine learning (ML) models, including artificial neural networks (ANNs) and random forests (RFs), resulted in 100% classification accuracy in the identification of RJ on the basis of feeding sources and floral periods, utilizing calculated fractionation factors. These findings demonstrate that δ¹³C and δ¹⁵N are reliable markers for RJ authenticity and highlight the importance of integrating isotopic data with feeding conditions for precise identification. The success of ANN and RF models underscores the potential of combining isotope fractionation with ML for high-precision traceability, offering a framework for food traceability and sustainability in apiculture.