Multivariate Analysis Improves Product Design Efficiency
September 1, 1996
Multivariate Analysis Improves By: Ray Marsili Food technology is a demanding profession. Food product designers are being asked to develop innovative new products that are high in quality but require minimum costs to produce - often using ingredients and/or processing technologies that didn't even exist a few years ago. Furthermore, industry-wide downsizing and competitive pressures mean that developing new products has to be accomplished more efficiently and in less time than ever before. Another ever-increasing challenge is quality control. Supermarket shelves are crowded with competing products, and new ones are waiting in line to replace them. Food processors can't afford to put inferior products on the market. Finished products must be monitored to ensure that optimum processing parameters are being consistently maintained; ingredient quality is high (e.g., no adulterated ingredients are being used); microbiological contamination hasn't occurred; and product color, taste, aroma and other attributes are acceptable. What tests should be performed as part of a meaningful quality control program? How are meaningful product specifications determined? The challenge facing food technologists isn't how to generate more objective data. Generating data is easy. Today's sophisticated, easy-to use, automated laboratory instruments can quickly generate more than enough data. But more data isn't necessarily better. Collecting more data doesn't mean your chances of solving problems are improved. What's needed is a better understanding of what the data really means - that is, a tool to find significant meaning in the mountains of data we already have. Clearly, food technologists need help with their product development tasks and quality control monitoring. One tool an increasing number of food scientists are turning to in order to help them get the most useful information from their data is multivariate analysis. It has proven to be one way to work smarter, faster and more efficiently. Univariate vs. multivariate Univariate measurements, which often are straightforward direct properties of a sample, can be a useful source of information. For example, weighing a filled bottle of ketchup is a useful univariate measurement for determining if fill weights agree with label claims. In many cases, however, simple univariate measurements don't provide enough information to resolve the problem at hand. Consider the example of a food chemist trying to determine why a sample of vanilla ice cream has an off-flavor. The food technologist may start an investigation by submitting the sample for chemical analysis. If chromatographic testing shows that vanillin and/ or other components of the vanilla extract are present in unusual amounts, the food technologist might assume the off-flavor problem may have been caused by the addition of an adulterated vanilla extract, or perhaps the wrong amount of vanilla extract was added to the product. But these conclusions may be erroneous. The cause may be butterfat oxidation products, microbial spoilage, overheating during processing, or perhaps use of the wrong type of corn syrup in the product formulation. Alternatively, the problem may have resulted from more than one of these possibilities, or from an interaction of two or more of these possibilities. Multivariate analysis (MVA) is superior to univariate analysis because, like the real world, it considers multiple variables simultaneously. For example, chromatographic analytical data can be examined with processing parameters, sensory analysis data, and market survey results to reveal variable relationships and data patterns that may not be obvious through a simple visual inspection of the data. The basics of MVA MVA is a family of statistical methods that uses multiple response measures on a series of objects or products. Measurements may be physical/chemical, process control, microbiological, sensory, or consumer-generated measurements. The goal of MVA is to develop a model that characterizes some property that is difficult to measure directly. The property may be a category classification (e.g., chemotyping essential oils, "good" flavor or "bad" flavor, the geographical origin of a food ingredient), or it may be a continuous property (e.g., product shelf life, or the concentration level of a critical ingredient). Initially, exploratory data analysis is performed on the data set. This step involves the computation and graphical display of patterns of association in multivariate data sets. Often the exploratory phase is accomplished with two techniques: Hierarchical Cluster Analysis (HCA) and Principal Component Analysis (PCA). Both can be used to show clustering and to reveal out-of-spec results. Classification modeling can be conducted, which involves the computation and graphical display of class assignments based on the multivariate similarity of one sample to others. One example of this technique is identification of bacteria based on fatty-acid profiles of lipid extracts from the bacterial cell walls. Two different classification models that are commonly used are K-Nearest Neighbor (KNN) and Soft Independent Modeling of Class Analogy (SIMCA). In addition, regression techniques such as Principal Component Regression (PCR) and Partial Least Squares (PLS) can be used to measure the degree of predictability. With regression models, the analyst is interested in predicting some value (rather than assigning a class designation) for an unknown sample. MVA in product development "I became a believer in the power of MVA several years ago when I was working on developing a new snack food," says David Lundahal, Ph.D., a consultant with Louisville, KY-based InfoSense, which specializes in offering training courses that teach MVA applications. The snack food was composed of several different ingredients. The processing, which was extrusion-based, involved many complicated steps. Dozens of experimental variables had to be investigated and optimized. Analytical testing and sensory taste paneling were conducted to evaluate finished product acceptance. Approximately 30 different samples were prepared (some with intentional defects), and over 100 variables (analytical data, sensory scores and processing parameters) were measured. One important goal of the study, says Lundahal, was to learn what sensory attributes were driving acceptance - and what attributes were responsible for unacceptance. Using various MVA techniques, Lundahal was able to determine the optimum processing conditions to use in order to manufacture a new snack food product with high consumer acceptance. Further, his studies revealed which sensory attributes were most critical to consumer acceptance and what chemical measurements were directly related to those attributes. Amazingly, he accomplished all this in less than six weeks - "an impossible accomplishment without the aid of MVA," he says. Not only did MVA allow Lundahal to develop a new product in record time, but it also provided him with insights into what type of physical and chemical QC testing to perform in order to ensure that acceptable products would be consistently produced, allowing the snack producer to set up a truly meaningful QC monitoring program. Another strong proponent of MVA for food product design is Tetsuo Aishima, Ph.D., manager of food science chemometrics, Research and Development Division, Kikkoman Corp. Aishima has been studying gas chromatography (GC) profiles of soy sauce for over 20 years. However, his analytical studies of soy sauce became significantly more useful after he started analyzing GC data and sensory data of samples with various MVA and chemometrics techniques. "Combining sensory data with analytical instrument data is a powerful tool when chemometric analysis is used," says Aishima, who uses the information to compare competitor products to Kikkoman's products and to fine tune Kikkoman's formulas for optimum consumer sensory preference. According to Aishima, a sister company of Kikkoman uses MVA to study tomato ketchup, tomato juice, and other types of tomato products. MVA as a QC tool Usually, measuring only one variable (e.g., pH) of a food product does not provide enough information to assess the product's flavor attributes; to determine if any processing changes have occurred that will lead to quality degradation (e.g., a shortened shelf life); and so on. By examining a series of parameters simultaneously with MVA, data generated by modern sophisticated analytical instruments - for example, near-infrared (NIR), Fourier transform infrared (FTIR), "electronic nose" instruments, etc. - can be significantly more revealing and useful than data obtained by traditional manual spot checks. Instrumental MVA techniques can provide various types of indications that warn of processing inconsistencies or out-of-specification ingredients. For the most part, a change in production processing parameters (e.g., cooking temperature or cooking time) will result in chemical changes in the product. On-line FTIR and NIR instruments can be used to detect these changes quickly, and analysis of the data by MVA techniques can be used to alert line operators as to exactly what processing parameters are out of control. This on-line approach provides practical and timely process control. It eliminates the need for lengthy, error-prone sample-preparation steps by lab personnel using conventional instrumental techniques. The results are speedier and more accurate. Negligible interruption of manufacturing production is another major benefit. MVA can be used to determine the geographical origin of food ingredients and raw materials (e.g., vanilla extract and spices). Geographical origin can determine the quality, acceptability and cost of a particular food or ingredient. MVA has been used to determine:
The technique also has been used to confirm food adulteration. For example, adulteration of orange juice can be detected using MVA by examining the concentrations of trace minerals (obtained by inductively coupled plasma analysis) and organic flavor (orange oil) components by capillary gas chromatography. Examining sugars and non-volatile acid concentrations by MVA has proven to be an effective way to detect apple juice adulteration. MVA has been used to determine when less-expensive rockfish is fraudulently being sold as premium priced red snapper. The fat from the fish is extracted, saponified and esterified. The fatty acid methyl esters (FAME) are then profiled by capillary GC. Data from authentic red snapper and rockfish are used to establish a classification model. Unknown samples can then be analyzed and compared to the model to detect mislabeling. Force behind lab instruments "Electronic nose" instruments have recently become commercially available, offering some exciting application potential for the food industry. For example, instruments have been used to verify the authenticity of Parmesan cheese; to aid in the control of the roasting and blending processes of coffee beans; to monitor the freshness of fruit when it is harvested, during shipment, and at the point of sale; to check the freshness of snack products, such as potato chips, to optimize shelf life; to characterize vegetable oils and determine purity; and to serve as quality assurance checks on expensive flavoring materials. Designing a chemical-sensor based instrument to quantitate all the individual volatile chemicals in a complex mixture of food aroma chemicals is impractical. The metal oxide-based and polymeric-based sensors, the two most commonly used types of sensors in today's instruments, have poor selectivity - i.e., they respond to many different types of chemicals rather than one individual chemical. An array of sensors is used, with each sensor responding to volatile organics in a slightly different way. By using MVA, sensor voltage responses can be tracked and transformed into useful information in the form of various pattern-recognition graphical displays, allowing for comparison of unknown samples to known calibration samples. Without the use of MVA, output from sensors would be impossible to interpret. Spectroscopic analysis also has benefited greatly in recent years from MVA applications. For example, recently introduced FTIR instruments specifically designed for food analysis can more accurately analyze for more different types of analytes because the instruments have incorporated MVA statistical software in their calibration protocols. Foss Food Technology Corp., Eden Prairie, MN, recently introduced the Milkoscan FT120, a new generation FTIR instrument for the analysis of dairy products. Using MVA statistics, the technique can distinguish subtle changes in FTIR absorption patterns when samples are analyzed and relate those changes to chemical composition changes in the food matrix. For example, the instruments can now detect and quantitate not only total carbohydrates, but also specific carbohydrates (e.g., lactose, glucose and fructose). Research is being conducted to determine if these new FTIR instruments can quantitate individual protein fractions in foods, as well - for example, casein in dairy products in addition to total protein. NIR instruments, which have never quite lived up to their potential, are now making a comeback thanks to improved MVA-based calibration techniques. Nearly every aspect of food industry R&D and QC can benefit from the use of MVA. Therefore, it is not surprising to see it being used more and more. We can expect most food technologists and food scientists to be using it as a routine tool in the near future.
© 1996 by Weeks Publishing Company 3400 Dundee Rd. Suite #100 |
You May Also Like