RESUMO
FTIR spectroscopy and multivariate analysis were used in the chemical study of the terroirs of Coffea canephora. Conilon coffees from Espírito Santo and Amazon robusta from Matas of Rondônia, were separated by PCA, with lipids and caffeine being the markers responsible for the separation. Coffees from Bahia, Minas Gerais, and São Paulo did not exhibit separation, indicating that the botanical variety had a greater effect on the terroir than geographic origin. Thus, the genetic factor was investigated considering the conilon and robusta botanical varieties. This last group was composed of hybrid robusta and apoatã. The DD-SIMCA favored the identification of the genetic predominance of the samples. PLS-DA had a high classification performance regarding the conilon, hybrid robusta, and apoatã genetic nature. Lipids, caffeine, chlorogenic acids, quinic acid, trigonelline, proteins, amino acids, and carbohydrates were identified as chemical markers that discriminated the genetic groups.
Assuntos
Coffea , Coffea/genética , Coffea/química , Cafeína/análise , Brasil , Café/química , LipídeosRESUMO
Considering the great economic significance of Coffea arabica (arabica) associated with the lower production cost of C. canephora (conilon), blends of these coffees are commercially available to reduce costs and combine sensory attributes. Thus, analytical tools are required to ensure consistency between real and labeled compositions. In this sense, chromatographic methods based on volatile analysis using static headspace-gas chromatography-mass spectrometry (SHS-GC-MS) and Fourier transform infrared (FTIR) spectroscopy associated with chemometric tools were proposed for the identification and quantification of arabica and conilon blends. The peak integration from the total ion chromatogram (TIC) and extracted ion chromatogram (EIC) was compared in multivariate and univariate scenarios. The optimized partial least squares (PLS) models with uninformative variable elimination (UVE) and chromatographic data (TIC and EIC) have similar accuracy according to a randomized test, with prediction errors between 3.3% and 4.7% and Rp2 > 0.98. There was no difference between the univariate models for the TIC and EIC, but the FTIR model presented a lower performance than GC-MS. The multivariate and univariate models based on chromatographic data had similar accuracy. For the classification models, the FTIR, TIC, and EIC data presented accuracies from 96% to 100% and error rates from 0% to 5%. Multivariate and univariate analyses combined with chromatographic and spectroscopic data allow the investigation of coffee blends.
Assuntos
Coffea , Coffea/química , Cromatografia Gasosa-Espectrometria de Massas , Café/química , Análise dos Mínimos Quadrados , Espectroscopia de Infravermelho com Transformada de FourierRESUMO
The chemical complexity of coffee influences the sensory evaluation of the beverage, the main method used to define the quality of the coffee. In view of the subjectivity that method offers, we propose the association of an instrumental method with multivariate calibration (PLS and GA-SVR) to predict the quality of arabica coffee as support for sensory analysis. Arabica coffee samples were submitted to sensory evaluation using the Specialty Coffee Association (SCA) protocol and HS-SPME-GC/MS analysis. The models presented RMSEp results from 0.20 to 0.25, within the evaluation range the quality levels of sensory attributes (0.25). For the fragrance/aroma attribute, a value of R2p equal to 0.8503 was reached. 15 volatile compounds were identified as responsible for predicting the quality of arabica coffee, among which, 1-nonadecene was first reported as an impact compound in the prediction of important sensory attributes.