INTERACTIVE SELF-MODELING MULTIVARIATE-ANALYSIS

Windig, W., Lippert, J.L., Robbins, M.J., Kresinske, K.R., Twist, J.P. and Snyder, A.P., 1990. Interactive self-modeling multivariate analysis. Chemometrics and Intelligent Laboratory Systems, 9: 7-30. In many practical applications in a laboratory, the data output from the analytical instruments is very complex. For example, a series of spectra following a reaction in time may contain information on intermediate and/or new products, of which no model spectra are available. In these kinds of cases, such standard data analysis tools as librrary search or subtraction methods cannot be applied. In order to analyze this type of data properly, self-modeling multivariate data analysis techniques have been developed. These techniques are capable of extracting spectra of the pure components from a data set of mixtures, without using prior knowledge of the pure components. This paper will explain a new approach for self-modeling multivariate analysis. The techniques involved will be explained by geometrical means. Examples of analyses of the data resulting from monitoring reactions in time will be shown. The first data set consist of Raman spectra monitoring the formation of silica glasses from a solution. The second data set consists of Fourier transform infrared data monitoring a reaction that produces a widely used ingredient in photography. The data have been analyzed with the Interactive Self-modeling Multivariate Analysis (ISMA) package, which is a higly interactive graphics oriented set of programs developed for this approach. © 1990.