PATTERN RECOGNITION FOR FOODS AND SUPPLEMENTS
2010 Annual Report
1a.Objectives (from AD-416)
To assist in the application of analysis of variance-principal components analysis (ANOVA-PCA) to food and botanical data acquired at ARS and to develop an algorithm for aligning chromatograms obtained with ultra-violet (UV) and mass spectrometric (MS) detection. The purpose of this cooperative agreement is to develop spectral fingerprinting and chromatographic profiling methods for the rapid detection and categorization of foods and botanical materials.
1b.Approach (from AD-416)
The Food Composition and Methods Development Laboratory (FCMDL) is responsible for developing analytical methods for chemical components in foods and dietary supplements. New methods are being developed for spectral fingerprinting and chromatographic profiling of plant materials using UV and MS detection. Fingerprints are acquired by direct analysis with no chromatographic separation. Both types of data are analyzed using pattern recognition programs like ANOVA-PCA. Proper interpretation of the data requires considerable skill in the area of chemometrics, the area of expertise of the cooperator. A particularly troublesome problem is that the application of ANOVA-PCA to chromatographic profiles requires the alignment of the chromatograms prior to processing. The cooperator will assist in the application of ANOVA-PCA to data obtained at USDA and in the analysis of variable interaction. In addition, the cooperators will develop software for aligning chromatograms so that ANOVA-PCA can be applied to chromatographic profiles.
In 2010, we used spectral fingerprinting and pattern recognition programs to examine 3 species of ginseng roots (Panax ginseng aka “Chinese” ginseng, P. quinquefolius aka “American” ginseng, and P. notoginseng). More than 50 samples of the 3 species were obtained from the Ginseng Board of Wisconsin, American Herbal Pharmacopoeia, and several commercial sources (some authentic roots directly from China). Mass spectrometry (MS) fingerprints acquired by direct analysis (root extracts with no chromatographic separation) readily identified the 3 species and the “red” and “white” preparations of the Chinese ginseng. Similar discrimination between species and preparations were obtained using UV absorption spectrophotometry and near infrared spectrometry. The latter method could discriminate between American ginseng grown in Wisconsin, Canada, and China. A follow-up study with 40 American ginseng roots from Wisconsin and China showed that MS could also discriminate between growing locations.
The computer programs used for pattern recognition were developed in collaboration with Ohio University. These programs consisted of soft independent modeling of class analogy (SIMCA), partial least squares-discriminant analysis (PLS-DA) and Fuzzy Rule-Building Expert Systems (FuRES). While all the methods reduce the large number of variables to (usually) 2 or 3 uncorrelated components, each uses a slightly different approach. Validation of the models was achieved by randomly dividing the samples between the “training set” and the “unknown samples”, building a model, and evaluating the accuracy of the model for the unknown samples. This was then repeated many times with a different, random distribution of samples each time.
This agreement was monitored through routine conversations and e-mails with the Ohio University.