Gas Chromatography/Olfactometry and Electronic Nose Analyses of Retronasal Aroma of Espresso and Correlation with Sensory Evaluation by an Artificial Neural Network

• Published in: Journal of food science 2010-11, Vol.75 (9), p.S477-S489
• Main Authors: Michishita, Tomomi, Akiyama, Masayuki, Hirano, Yuta, Ikeda, Michio, Sagara, Yasuyuki, Araki, Tetsuya
• Format: Article
• Language: English
• Subjects: Aroma; artificial neural network (ANN); Biological and medical sciences; Brazil; CharmAnalysis; Chromatography; Coffee; Coffee, tea and other stimulative beverage industries; Colombia; Correlation analysis; Discriminant Analysis; electronic nose (E-nose; Electronic noses; Electronics; Ethiopia; Food industries; Food science; Fundamental and applied biological sciences. Psychology; Gas Chromatography-Mass Spectrometry - methods; gas chromatography/olfactometry (GC/O; Guatemala; Indonesia; Learning theory; Linear Models; Mathematical models; Neural networks; Neural Networks (Computer); Odorants - analysis; Odors; Olfactory Perception; retronasal aroma of espresso coffee; Roasting; sensory evaluation; Sensory perception; Tanzania; Taste Perception; Volatile Organic Compounds - analysis; αFOX4000
• ISSN: 0022-1147; 1750-3841
• DOI: 10.1111/j.1750-3841.2010.01828.x

:  To develop a method for evaluating and designing the retronasal aroma of espresso, sensory evaluation data was correlated with data obtained from gas chromatography/olfactometry (GC/O, CharmAnalysis™) and from an electronic nose system αFOX4000 (E‐nose). The volatile compounds of various kinds of espresso (arabica coffee beans from 6 production countries: Brazil, Ethiopia, Guatemala, Colombia, Indonesia, and Tanzania; 3 roasting degrees for each country: L values, 18, 23, and 26) were collected with a retronasal aroma simulator (RAS) and examined by GC/O and E‐nose. In addition, sensory descriptive analysis using a 7‐point scale for RAS effluent gas was performed by 5 trained flavorists using sensory descriptors selected based on the frequency in use and coefficient of correlation. The charm values of 10 odor descriptions obtained from GC/O analysis exhibited the significant (P < 0.05) differences among both roasting degrees and origins. Also, linear discriminant analysis (LDA) on the E‐nose‐sensor resistances and factor analysis on the sensory evaluation scores showed that the differences of aroma characteristics among the roasting degrees were larger than those among the origins. Based on an artificial neural network (ANN) model applied to the data from GC/O analyses and sensory evaluations, the perceptual factor of the RAS aroma was predicted to be mainly affected by sweet‐caramel, smoke‐roast, and acidic odors. Also, 3 metal oxide semiconductor sensors (LY2/Gh, P30/1, and T40/1) of E‐nose were selected for analyses of RAS aroma and correlated with the sensory descriptive scores by the ANN to support sensory evaluation.