Statistical analysis model discrimination

A total of 185 emission lines for both major and trace elements were attributed from each LIBS broadband spectrum. Then background-corrected, summed, and normalized intensities were calculated for 18 selected emission lines and 153 emission line ratios were generated. Finally, the summed intensities and ratios were used as input variables to multivariate statistical chemometric models. A total of 3100 spectra were used to generate Partial Least Squares Discriminant Analysis (PLS-DA) models and test sets. 

For a given rate expression this yields the optimal parameter values, but it has to be decided whether the rate expression is the most adequate one. This selection can be based on statistical analysis and on physical significance of the parameter values (for example they should be positive in many cases). Additional experimental effort can be used towards further model discrimination by carefully planning experiments - a kind of experimental design method. 

It is commonly the case that a wide variety of properties can be included in a QSAR analysis and a decision must be made on whether to include all possibilities or limit the number of descriptors. This decision depends on the size of the data set and the correlation matrix between the properties. Farge sets of property data contain a lot of redundancy of information. For example, molecular weight, surface area and molar refraction are always highly correlated, therefore a decision to nse only molecular weight could be made. Some multivariate statistical analysis methods are tolerant of data sets which contain more properties than compounds, for example, PFS, while others are not, for example, linear discriminant analysis (FDA). Ideally, a set of uncorrelated properties is desirable as this is most likely to give a robust, interpretable model. 

The problems of parameter estimation and model discrimination by statistical analysis are most easily understood in relation to a specific example. Let us consider a bisubstrale reaction that obeys the rate Eq. (18.50). The initial rate data from Table 1 are presented graphically in Figs. 3 and 4. The reciprocal plot of l/Uo versus 1/A is a family of straight lines which, in this case, may look either parallel or intersecting due to the fact that the crossover point is distant from the vertical axis, because If is much smaller than K, it would be... 

Graphical analysis must always precede the statistical analysis. It is imperative to keep short the time elapsed between data acquisition and data analysis, and in most cases, it is advisable to perform the graphical analysis even while the experiment is still in progress. Wien the data clearly define the nature ofthe rate or binding equation, statistical analysis is not needed to do this. Nevertheless, for a definitive work, statistical methods are necessary for parameter estimation as well as for model discrimination (Senear Bolen, 1992). Computer programs are now available for even the most sophisticated problems in enzyme kinetics (see Section 18.2.4). 

The proper combination of graphical and statistical analysis will usually provide definitive answers for most problems in model discrimination. Severd specific examples are given below. 

Models for homologous series of substances using simple regression methods. Quantitative substructure-based models derived from multivariate statistics (e.g. discriminant analysis), which may be applicable to a variety of chemical classes. 

The VolSurf method was used to produce molecular descriptors, and PLS discriminant analysis (DA) was applied. The statistical model showed two significant latent variables after cross-validation. The 2D PLS score model offers a discrimination between the permeable and less permeable compounds. When the spectrum color is active (Fig. 17.2), red points refer to high permeability, whereas blue points indicate low permeability. There is a region in the central part of the plot with both red and blue compounds. In this region, and in between the two continuous lines, the permeability prediction is less reliable. The permeability model... 

Current methods for supervised pattern recognition are numerous. Typical linear methods are linear discriminant analysis (LDA) based on distance calculation, soft independent modeling of class analogy (SIMCA), which emphasizes similarities within a class, and PLS discriminant analysis (PLS-DA), which performs regression between spectra and class memberships. More advanced methods are based on nonlinear techniques, such as neural networks. Parametric versus nonparametric computations is a further distinction. In parametric techniques such as LDA, statistical parameters of normal sample distribution are used in the decision rules. Such restrictions do not influence nonparametric methods such as SIMCA, which perform more efficiently on NIR data collections. 

---