Species and geographical origin authentication of Philippine coffee using XRF-based multi-element and stable isotope ratio profiling combined with chemometric tools and machine learning algorithm

• Main Author: Tan, Kevin Neil G.
• Format: text
• Language: English
• Published: Animo Repository 2023
• Subjects: Coffee > Philippines; Chemistry
• Online Access: https://animorepository.dlsu.edu.ph/etdm_chem/17; https://animorepository.dlsu.edu.ph/context/etdm_chem/article/1017/viewcontent/2023_Tan_Species_and_Geographical_Origin_Authentication_of_Philippine_Coff_Full_text.pdf
• Institution: De La Salle University

Multi-element and stable isotope ratio (SIR) profiling with chemometrics and machine learning techniques can provide a means to differentiate roasted coffee beans based on their species (Arabica and Robusta) and geographical origin. This approach can help mitigate food fraud and secure the geographical indication (GI) of Philippine coffee. Cultivation practices, post-harvest processes, and environmental factors such as soil composition, precipitation, temperature, and altitude influence the chemical composition of a coffee bean. A total of fifty-six (56) roasted coffee bean samples were collected from the participants of the 2022 Philippine Coffee Quality Competition (PCQC). Eight (8) commercially available roasted coffee beans were also collected. XRF-based multi-element and stable isotope ratio profiles from these two sets of samples were subjected to principal component analysis (PCA), linear discriminant analysis (LDA), and random forest (RF). Samples are categorized based on regions. The concentrations of P, S, K, Ca, Mn, Fe, Cu, Zn, Rb, Sr, δ13C, and δ15N were utilized as predictors to differentiate and classify samples based on species and geographical origin. RF provides higher accuracy than LDA on species classification (98.21% vs. 94.64%). On the other hand, region-based geographical origin classification accuracy is higher in LDA than in RF (74.07% vs. 64.82%). Including SIRs (δ13C and δ15N) as explanatory variables for classification increases the accuracy of the LDA model by 3.70% and the RF model by 3.71% in geographical origin classification. Based on the generated models, δ13C is a better predictor than δ15N in discriminating coffee based on geographical origin. XRF-based multi-element profiling can be used for high-throughput screening of species and the geographical origin of coffee. XRF is a fast, cost-efficient, and reliable instrumental method for multi-elemental analysis. LDA and RF are viable statistical tools for utilizing XRF-based multi-element and SIR profiles to accurately classify coffee based on species and geographical origin.