Other pattern recognition approaches

When dealing with complex matrices such as those encountered in food and medical samples, it is often difficult to develop a sensor that exhibits a sufficiently selective response for the target analyte. Under these circumstances, arrays and mathematical modeling, for example, principal components analysis, artificial neural networks, or other pattern recognition approaches, may be required [42],... 

Although the coining of the term metabolomics has been relatively new, an explosion of publications has occurred in this field plus a realization that many researchers already were doing similar studies, albeit without an -omic tag before the word. It has not been possible to review all the applications of metabolomics fully, and the applications will increase. In addition to the development of new applications, the development of the analytical approaches will also take center stage as researchers push back the limits of detection of NMR spectroscopy, mass spectrometry, and other analytical approaches. In addition to these wet lab developments, both the pattern recognition approaches used to process metabolomics and the metabolomic databases used to identify metabolites need to be developed or expanded. In this respect, an excellent place to start on the arduous journey to biomarker discovery through metabolomics is the current metabolomic databases found on the web that make standard spectra freely available (68-70). 

This is a pattern recognition approach, which does not take into account individual congeners that might occur, such as in reaction by-products. This test method describes the use of Aroclors 1016, 1221, 1232, 1242, 1248, 1254, 1260, 1262, and 1268, as reference standards, but others could also be included. Aroclors 1016 and 1242 have similar capillary GC patterns. Interferences or weathering are especially problematic with Aroclors 1016, 1232, and 1242 and may make distinction between the three difficult. 

While principal components models are used mostly in an unsupervised or exploratory mode, models based on canonical variates are often applied in a supervisory way for the prediction of biological activities from chemical, physicochemical or other biological parameters. In this section we discuss briefly the methods of linear discriminant analysis (LDA) and canonical correlation analysis (CCA). Although there has been an early awareness of these methods in QSAR [7,50], they have not been widely accepted. More recently they have been superseded by the successful introduction of partial least squares analysis (PLS) in QSAR. Nevertheless, the early pattern recognition techniques have prepared the minds for the introduction of modem chemometric approaches. 

From both a theoretical and practical view, it is ideal to use Bayesian Decision Theory because it represents an optimal classifier. From a theoretical perspective, Bayesian Decision Theory offers a general definition of the pattern recognition problem and, with appropriate assumptions, it can be shown to be the basis of many of the so-called non-PDF approaches. In practice, however, it is typically treated as a separate method because it places strong data availability requirements for direct use compared to other approaches. 

Other approaches Hamming networks, pattern recognition, wavelets, and neural network learning systems are sometimes discussed but have not been commercially implemented. 

The two pattern recognition techniques used In this work are among those usually used for unsupervised learning. The results will be examined for the clusters which arise from the analysis of the data. On the other hand, the number of classes and a rule for assigning compounds to each had already been determined by the requirements of the mixture analysis problem. One might suppose that a supervised approach would be more suitable. In our case, this Is not so because our aim Is not to develop a classifier. Instead, we wish to examine the data base of FTIR spectra and the metric to see If they are adequate to help solve a more difficult problem, that of analyzing complex mixtures by class. 

For the pattern recognition (PARC) approach, we have coated the piezoresistive cantilevers with different selective layers. Each piezoresistive cantilever had four cantilever elements. Two of these cantilevers were coated with gold, whereas the other two served as reference cantilevers. We have used four separated chips arranged into an array in a single vapor chamber. Each cantilever chip was coated with a different selective agent. The four coatings used in our study include 4-MBA, Au (evaporated), CH3(CH2)n-SH, and a complex of (i-cyclodextrin and alkane. 

An important application of PCA is classification and pattern recognition. This particular application of PCA is described in detail in Chapter 9. The fundamental idea behind this approach is that data vectors representing objects in a high-dimensional space can be efficiently projected into a low-dimensional space by PCA and viewed graphically as scatter plots of PC scores. Objects that are similar to each other will tend to cluster in the score plots, whereas objects that are dissimilar will tend to be far apart. By efficient, we mean the PCA model must capture a large fraction of the variance in the data set, say 70% or more, in the first few principal components. 

In Chapters 4 and 5, we discussed a number of mediods for multivariate data analysis, but the methods described did not take into account the sequential nature of the information. When performing PCA on a data matrix, die order of die rows and columns is irrelevant. Figure 6.1 represents three cross-sections through a data matrix. The first could correspond to a chromatographic peak, the others not. However, since PCA and most other classical methods for pattern recognition would not distinguish these sequences, clearly other approaches are useful. 

Another significant environmental factor for vapor-phase applications is humidity. The ubiquitous nature of water vapor requires development of means to exclude or correct for interferences from water [92a,b]. Careful selection of coating materials, for example, can minimize the effect of water vapor on the sensor response. Alternatively, a coating with appropriate sensitivity to water can be used in the development of correction algorithms [93]. Other instrumental or system approaches, such as preconcentrators or sensor arrays with pattern recognition [94a-c], will be discussed in Section 5.5 and in Chapter 6. 

Recognition of Unusual Patterns. Recognition of unusual patterns requires a population-based or epidemiological approach to data analysis and interpretation. What would constitute an unusual pattern of disease occurrence Essentially, it is a cluster that does not fit. A cluster is an aggregation of cases of a disease or other health-related condition... which are closely grouped in time and place (CDC, 1992, p. 429). Again, baseline information is needed for comparison to evaluate what is unusual. 

Note that an integral part of the metabolomic approach is the application of pattern recognition techniques to deduce what variation in a data set is associated with a given disease, genetic modification, or other manipulation of the system. Because of space limitations, it is not possible to discuss this area as part of this article, but we refer the reader to several excellent reviews (33, 34). 

Other attempts have focused on proof-of-concept aspects of classification, using standard supervised pattern recognition methods. Major deficiencies exist in this approach when attempting the classification of biomedical spectra, including the following. 

---