Reliable discriminant analysis tool for controlling the roast degree of coffee samples through chemical markers approach

Roasting is one of the most influencing stages of coffee processing. Accordingly, a discriminant analysis (DA) was carried out with the objective of identifying key compounds (chemical markers) that enable a differentiation of coffee samples according to their roasting degree. For this, chromatographic data of the volatile fraction of 21 coffee samples submitted to distinct roasting treatments (Light, Medium, Dark, and French Roasts) were employed. Using three discriminant functions that rely on only ten chemical markers, it was possible to explain 100 % of the variance of the data points. If two functions are used, the surprisingly high value of 99.4 % is achieved. The model was cross-validated, and the main function successfully passed a permutation test using two statistical indicators. It was found that half of the markers belong to the pyrazines family, known to grant sensorial notes related to roasted hazelnut and peanuts. In the whole, this essay demonstrates the usefulness of DA as a tool to control the quality of roasting treatment of coffee and can be further extended with advantage to the eight roasting degrees of the AGTRON Roasting Classification as soon as larger databases become available.