Development of an odour mixture model for cheese flavour

Abstract

This research set out to examine the feasibility of developing a mathematical model that would allow the prediction of the sensory characteristics of mixtures of odour compounds.

Initially, a trained sensory panel used descriptive analysis to describe the characteristics of four cheese odour compounds individually. The compounds were diacetyl, ethyl butyrate, butyric acid, and methional. This preliminary work showed that the characteristics of an odour can change as its concentration is increased. No previous studies had examined the way in which concentration changes can affect the perceived quality of an odour.

The primary sensory tool used in this research was descriptive analysis. Because descriptive analysis requires panelists to be extensively trained, and to behave in an analytical way, it was important to gain some insights into the way in which the results obtained might be influenced by the methodology employed. To this end, binary odour mixtures were studied using an alternative, similarities-based task, with groups of both trained and untrained panelists. The series of experiments carried out demonstrated that the perception of binary odour mixtures was not affected by the training and experience of the panelists, nor by the task required of them, nor by the odours themselves.

The trained panel then evaluated selected mixtures of diacetyl and ethyl butyrate, using the attributes that they had generated in the initial stage of the work. The data from the panel's evaluation of the mixtures was subjected to response surface regression. In order to test the predictive ability of the resultant equations, the panel evaluated additional mixtures of the two compounds. This extra data was then combined with the main data and the response surface regression was repeated to ensure that the coefficients of the regression equations were not significantly altered by the inclusion of the additional data. This approach was repeated with mixtures of three compounds and then with mixtures of four compounds. The final result was a series of twelve regression equations that could be used to predict the sensory characteristics of any combination of the four odour compounds.

The model developed, which consisted of the twelve regression equations, was tested by applying it to the perception of Parmesan cheese. The panel evaluated three different Parmesan-type cheeses using the attributes that they had used to describe the odour mixtures. The regression equations were then used to predict combinations of the four odour compounds that would be likely to give the most similar sensory profile to the Parmesan cheeses. The panel then assessed the odour of these mixtures, and of a range of cheeses, including Parmesan and non-Parmesan types, to determine their similarity to the panelists' concept of Parmesan cheese odour. All of the odour mixtures were as or more similar to the panelists' concept of Parmesan cheese odour as the four Parmesan-type cheeses included in the experiment. The Cheddar and Gouda cheeses included in the experiment were less similar to the panelists' concept of Parmesan cheese odour than any of the odour mixtures or Parmesan cheeses. Thus it was concluded that a model had been developed that was capable of predicting the sensory characteristics of any combinaton of diacetyl, ethyl butyrate, butyric acid, and methional, and that this model could be used to predict combinations of these four odour compounds that will give rise to desired sensory characteristics in a formulated food product.

It is envisaged that this approach could be used extensively in the formulation of food products, or other formulated products in which the perception of odour is important. A model developed in this way could be used to predict what sensory characteristics would be likely to be perceived in a product formulated with a specific mixture of odour compounds, or to ascertain which combination of odour compounds is likely to give rise to desired sensory characteristics.