Analysis and classification of coffee beans using single coffee bean mass spectrometry with machine learning strategy

Abstract

Coffee is a daily essential, with prices varying based on taste, aroma, and chemical composition. However, distinguishing between different coffee beans is challenging due to time-consuming and destructive sample pretreatment. This study presents a novel approach for directly analyzing single coffee beans through mass spectrometry (MS) without the need for sample pretreatment. Using a single coffee bean deposited with a solvent droplet containing methanol and deionized water, we generated electrospray to extract the main species for MS analysis. Mass spectra of single coffee beans were obtained in just a few seconds. To showcase the effectiveness of the developed method, we used palm civet coffee beans (kopi luwak), one of the most expensive coffee types, as model samples. Our approach distinguished palm civet coffee beans from regular ones with high accuracy, sensitivity, and selectivity. Moreover, we employed a machine learning strategy to rapidly classify coffee beans based on their mass spectra, achieving 99.58% accuracy, 98.75% sensitivity, and 100% selectivity in cross-validation. Our study highlights the potential of combining the single-bean MS method with machine learning for the rapid and non-destructive classification of coffee beans. This approach can help to detect low-priced coffee beans mixed with high-priced ones, benefiting both consumers and the coffee industry.