Rapid Convolutional Algorithm for the Discovery of\nBlueberry Honey Authenticity Markers via Nontargeted LC-MS Analysis

Bees\nproduce honey through the collection and transformation of\nnectar, whose botanical origin impacts the taste, nutritional value,\nand, therefore, the market price of the resulting honey. This phenomenon\nhas led some to mislabel their honey so that it can be sold at a higher\nprice. Metabolomics has been gaining popularity in food authentication,\nbut rapid data mining algorithms are needed to facilitate the discovery\nof new authenticity markers. A nontargeted high-resolution liquid\nchromatography–mass spectrometry (HR/LC–MS) analysis\nof 262 monofloral honey samples, of which 50 were blueberry honey,\nwas performed. Data mining methods were demonstrated for the discovery\nof binary single-markers (compound was only detected in blueberry\nhoney), threshold single-markers (compound had the highest concentration\nin blueberry honey), and interval ratio-markers (the ratio of two\ncompounds was within a unique interval in blueberry honey). A novel\nconvolutional algorithm was developed for the discovery of interval\nratio-markers, which trained 14× faster and achieved a 0.2 Matthews\ncorrelation coefficient (MCC) units higher classification score than\nexisting open-source implementations. The convolutional algorithm\nalso had classification performance similar to that of a brute-force\nsearch but trained 1521× faster. A pipeline for shortlisting\ncandidate authenticity markers from the LC-MS spectra that may be\nsuitable for chemical structure identification was also demonstrated\nand led to the identification of niacin as a blueberry honey threshold\nsingle-marker. This work demonstrates an end-to-end approach to mine\nthe honey metabolome for novel authenticity markers and can readily\nbe applied to other types of food and analytical chemistry instruments.