This paper presents a study that assesses the application of chemometrics for classifying coffee samples in a quality control context. High-resolution and accurate mass measurements were utilized as input for pixel-based orthogonal partial least squares discriminant analysis (OPLS-DA) models. The compositional data were acquired through a fully automated workflow combining headspace solid-phase microextraction and gas chromatography-high-resolution mass spectrometry (GC-HRMS) using an FT-Orbitrap® mass analyzer. A workflow centered on accurate mass measurements was successfully utilized for group-type analysis, offering an alternative to methods relying solely on MS similarity searches. The predictive models underwent thorough evaluation, demonstrating robust multivariate classification performance. Five key coffee attributes, bitterness, acidity, body, intensity, and roasting level were successfully predicted using GC-HRMS data. The results revealed strong predictive accuracy across all models, ranging from 88.9 % (bitterness) to 94.4 % (roasting level). This study represents a significant advancement in automating methods for coffee quality control, notably increasing the predictive ability of the models compared to existing literature.
Read full abstract