Enhancing chromatographic classification of tea varieties using an optimised gradient and peak alignment
Year: 2025
Type: article
Abstract: • 51 brand/raw materials, 25 extra replicates and 21 alignment samples were processed. • Direct signals required sacrificing information by defining region bands. • Prior classification was hindered by peak shifts, limited representation and outliers. • Alignment with periodic references allowed processing full signals preserving richness. • PLS-DA enabled visualising the expected 6 tea varieties, improving clarity and accuracy. Medicinal plants synthesise bioactive molecules that play a significant role in disease treatment and contribute to the development of pharmaceutical products, driving advancements in modern Medicine and therapeutic strategies. Chromatographic fingerprints, which capture the diversity of chemical constituents of these plants, are essential for their authentication and classification. However, traditional optimisation methods, which primarily focus on characterising or identifying individual components, are often insufficient to achieve optimal chromatographic profiles for complex natural products. To address this challenge, previous work employed a global retention model that considered all detected peaks within a sample to optimise the gradient program. This approach was applied to classify samples from tea varieties, including green, black, white, red, oolong, and rooibos. The dataset, however, had limited representation of rooibos and oolong categories. Also, alignment of chromatograms was needed to reduce signal fluctuations and improve data quality. For this purpose, in this work, Correlation Optimised Warping (COW) was used, facilitated by the periodic injection of a reference mixture. This alignment enabled the processing of baseline-corrected signals, significantly enhancing the classification accuracy among tea varieties. To assess classification feasibility, Principal Component Analysis (PCA), k –means clustering, and Partial Least Squares-Discriminant Analysis (PLS-DA) were employed. These methods provided a robust framework for exploring the distinct chemical profiles of the tea samples. Applying PLS-DA to a new tea dataset comprising 51 samples from commercial brands / bulk raw materials, 25 additional replicates, and 21 alignment samples effectively differentiated tea varieties with improved reliability and accuracy compared to previous studies.
Source: Microchemical Journal
Institution: Universitat de València
Language: English
Subfield: Analytical Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,000
Award: G3194991932
OpenAlex ID: W4409264520