Ribeiro J. S., Salva T. J. G., Tomaziello R. A., Augusto F., Ferreira M. M. C., "PREDICTION OF SENSORY PROFILES OF BRAZILIAN ARABICA COFFEE USING AROMA COMPOSITION, GAS CHROMATOGRAPHY AND PARTIAL LEAST SQUARES REGRESSION MODELS". Campinas, SP, Brazil, 14-19/09/2008: 22nd International Conference of Coffee Science, Programme & Abstracts (2008) 38. Oral C318.

RIBEIRO, Juliano S.*, SALVA, Terezinha J. G**, TOMAZIELLO, Roberto A.**, AUGUSTO, F.* and FERREIRA, Márcia M. C*

*Universidade Estadual de Campinas, SP, Brazil, **Instituto Agronômico de Campinas, SP, Brazil.

The establishment  of  mechanisms that allow the evaluation,  assurance  and  quality certification of  food  products  is  an
indispensable strategy for maintaining commercial competitiveness. Sensory profiling is still the most widespread technique
employed  to  evaluate the final quality of coffee.   The correlation between flavour  and  sensory profiles using  multivariate
analysis becomes an excellent tool  in  the quality control of foods  and  agricultural products,  being applied successfully in
analyses of hazelnut, vinegar, juices and wine.

In this work,   the volatile compounds  of  fifty-eight Brazilian arabica roasted coffee samples were analyzed  by solid phase
microextraction  and  gas chromatography  (SPME-GC)  and  the  data were used  to  correlate with sensory evaluation by
experts. The sensory attributes investigated were body, flavour, cleanliness and overall quality. Regression models (partial
least squares) [1] based on correlation of chromatographic profiles with each sensory attribute were developed (Figure 1A).
The ordered predictor selection (OPS) method was used for variable selection [2].
 

Figure 1 – Predictions (A) and Hotelling T2 (B) of calibration (circles) and prediction samples (triangles).
 
 

For the calculation of the prediction errors, Hotelling T² statistics had been used [3]. The results provided by the regression
models are promising and prove the feasibility of using  a similar methodology in routine applications  to predict the sensory profiles of unknown Arabica coffee samples.

[1] Ferreira, M. M. C., Antunes, A. M., Melgo, M. S., Quim. Nova, 22, 724-731 (1999);

[2] Teófilo, R. F., Martins, J. P., Ferreira, M. M. C., register 0000270703255138: Brazil (2007);

[3] Hotelling, H., J. Educ. Psychol., 24, 417-441 (1933);
 
 

38