Detection of Rancid Defect in Virgin Olive Oil by the Electronic Nose Ramo ´n Aparicio,* ,† Silvia M. Rocha, ‡ Ivonne Delgadillo, ‡ and Marı ´a Teresa Morales † Instituto de la Grasa (CSIC), Avenida Padre Garcı ´a Tejero 4, 41012 Sevilla, Spain, and Departamento de Quı ´mica, Universidade de Aveiro, 3810 Aveiro, Portugal A sensor array of 32 conducting polymer sensors has been used to detect the rancid defect in virgin olive oils. A training set, composed of admixtures of a Portuguese virgin olive oil with different percentages (0-100%) of a rancid standard oil, was used for the selection of the best sensors classifying correctly the samples. Information on volatile compounds responsible for rancidity and the sensory evaluation of samples by assessors were used for explaining the mathematical selection of sensors. A tentative calibration, using unsupervised procedures (PCA and MDS) and a nonlinear regression, was carried out, with the training set, and later confirmed with a test set with which rancid commercial samples of different varieties were used to spike a Greek extra virgin olive oil at low levels of rancidity (0.5-6%). Keywords: Virgin olive oil; electronic nose; sensory evaluation; volatiles; rancidity INTRODUCTION Virgin olive oil sensory quality is currently deter- mined by the European Union regulation (EC, 1991) or the International Olive Oil Council trade standards (IOOC, 1996). Both official methods carry out the sensory evaluation by using panels of trained assessors, although there are certain technical differences between them, basically the kind of sensory descriptors and the scale of evaluation (structured versus nonstructured). These panel tests have been useful for the construction of a consensus between experts of different countries about which attributes would be present or absent in virgin olive oils of reputable quality. However, panel tests are a costly and slow procedure that is not always at the disposal of small producers or cooperative societ- ies; only large retailers and sellers may be able to afford such tests. Furthermore, the subjective opinion of as- sessors undermines the final overall evaluation, and some flaws have been pointed out (Ranzani, 1994), mostly when multivariate procedures are not applied (Aparicio et al, 1992). On the other hand, volatile compounds are respon- sible for the flavor perceptions detected by assessors, and they are not obviously subjective information but quite objective (Flath et al., 1973). On the basis of this fact, a recently proposed methodology correlates basic sensory descriptors (Aparicio et al., 1994; Aparicio and Morales, 1995) with the volatile compounds responsible for them (Morales et al., 1995; Aparicio et al., 1996). The methodology, to which the mathematical procedures give support (Aparicio and Morales, 1994; Morales et al., 1994), has allowed the explanation of the most remarkable virgin olive oil sensory descriptors (green, bitter-pungent-astringent, sweet, fruity, ripe fruit, ripe olives, and miscellaneous undesirable attributes). However, this methodology requires quantification of the volatile compounds by dynamic headspace high- resolution gas chromatography (DHS-HRGC), which is today slow enough to be applied to on-line processes. The on-line control is a demand more and more heard from producers who want to store their olive oils in different deposits, according to olive oil quality, as soon as the oils have been produced from the automatic centrifugation systems. The importance is not then in the detection of those attributes responsible for high- quality virgin olive oils, for example, green and fruity (IOOC, 1996), but in the quantification of defects (Peri and Rastelli, 1994), rancidity being one of the most remarkable undesirable attributes. As lipids oxidize, they form hydroperoxides, which are susceptible to further oxidation or decomposition to secondary reaction products such as aldehydes, ketones, acids, and alcohols. In almost all cases, these compounds adversely affect flavor, aroma, taste, nutritional value, and overall quality (Vercelloti et al., 1992). Because there are many catalytic systems that can oxidize lipids, for example, light, temperature, enzymes, metals, met- alloproteins, and microorganisms, the control, quanti- fication, and prediction of oxidation are still important issues from either scientific or economic points of view. From a chemical viewpoint, different methodologies have been suggested to measure and predict the oxida- tion (Gutie ´rrez, 1989; ISO 6886, 1989; Cabre ´ and Masso ´, 1992; Morales and Aparicio, 1997). From a sensory point of view, a virgin olive oil is oxidized when assessors detect and quantify the presence of the rancid percep- tion in the complex matrix of virgin olive oil (EC, 1991; IOOC, 1996). A potential alternative, based on semiconductive organic polymers, has recently appeared for the evalu- ation of foods and food products (Persaud, 1991; Taylor et al., 1995; Dube and Peterson-Daly, 1996; Taylor, 1998; Tullett, 1996; Goldring, 1997; Visser and Taylor, 1998), although certain restrictions have been detected in their applications (Zannoni, 1995; van Ysacker and * Author to whom correspondence should be addressed (fax +34-95-4616790; e-mail aparicio@cica.es). † Instituto de la Grasa (CSIC). ‡ Universidade de Aveiro. 853 J. Agric. Food Chem. 2000, 48, 853-860 10.1021/jf9814087 CCC: $19.00 © 2000 American Chemical Society Published on Web 02/18/2000