Electronic nose can detect light-oxidized off-flavors in reduced-fat milk.

The electronic nose, an array of gas sensors and pattern recognition software, is able to rapidly and nondestructively analyze odors and flavors in foods. It may be possible to use this device in the headspace analysis of dairy products. Here it would help determine off-flavors caused by light-induced product deterioration.

Light-oxidized off-flavors in milk occur when the product is exposed to fluorescent light in retail dairy cases. The degradation of sulfur-containing amino acids creates burnt feather-like off-flavors from low-molecular-weight sulfur volatiles. Moreover, the oxidation of unsaturated fatty acids generates cardboard-like, metallic off-flavors from aldehydes and ketones.

Scientists at Michigan State University found that an electronic nose can detect light-oxidized off-flavors in reduced-fat milk, offering comparable sensitivity to triangle sensory tests. In their research, the scientists stored reduced fat 2% milk in 237-ml glass bottles at 5 C. These were exposed to up to 48 hours of fluorescent light. The investigators used an electronic nose equipped with 12 metal-oxide semiconductor sensors to analyze the headspace generated at 45 C, 70 C or 95 C for 15 minutes. Models based on sensor responses using linear or quadratic discriminant function analysis (LDA or QDA) or k-nearest neighbor (k-NN) techniques were evaluated.

Two dozen consumer panelists performed a series of triangle tests to determine the sensory threshold of light-oxidized off-flavors. The tests showed that light caused significant olfactory changes in the product during an eight-hour period. Canonical discrimination scatter plots indicated a partial differentiation in LDA models based on the sensor responses of the electronic nose.

A headspace temperature of 95 C offered better differentiation for moderately oxidized samples taken at up to 12 hours, although a 45 C headspace environment differentiated highly oxidized samples taken at 36 hours and 48 hours. Among several supervised statistical learning techniques, LDA and 1-NN had the highest correct identification rates for the test data. When harmful light exposure was defined at eight hours, the LDA of 95 C headspace samples correctly recognized 97% of the oxidized samples.

Further information. John Partridge, Department of Food Science and Human Nutrition, Michigan State University, 2100B Anthony Hall, East Lansing, MI 48824; phone: [phone]; fax: [phone]; email: [e-mail].