Botanical Identification: A Chemometric Challenge

Conceptually, identification, or authentication, of a botanical material is simple. You collect authentic materials, make a series of analytical measurements, build a model, set statistical limits, and then determine whether the unknown test material fits the model. The challenges to implementing such a method are 1) collecting authentic materials that reflect the expected natural variance and 2) choosing the appropriate analytical and data processing methods to characterize the samples as thoroughly as possible. Ultimately, the method must produce a single value that can be compared to a predetermined threshold and give a YES or No answer. The identification of botanical materials is more robust if more data are used. Hence the full chromatograms, from GC, LC, CE, and TLC, or full spectra, from MS, NMR, IR, NIR, and UV, have been used with the various multivariate methods. Chemometric methods are well suited for treatment of chromatograms and spectra as images that can be used to construct an authentic model. The appropriate method is dependent of whether the goal is identification or classification and whether the method is targeted or non-targeted. Soft independent modeling of class analogy (SIMCA) models each class of samples and is ideally suited for identification where the only available information is the characteristics of the authentic materials. Partial least squares-discriminant analysis (PLS-DA) is better suited for classification, where the characters of 2 or more materials are known. In both cases, botanical identification is a probability problem and the analyst can control the statistical limits.