REDUCING DIMENSIONALITY IN PRINCIPAL COMPONENT ANALYSIS. A METHOD COMPARISON

Zsuzsa Kánya, Esther Forgács and Tibor Cserháti

Institute of Chemistry, Chemical Research Center, Hungarian Academy of Sciences,
P.O. Box 17, 1525 Budapest, Hungary

Zoltán Illés

Central European University, Budapest, Hungary

The retention time of l0 ring-substituted analine derivatives has been measured on 5 different reversed-phase chromatographic HPLC columns using various mixtures of water and methanol as mobile phases. The retention parameters (log k’, theoretical plate number and asymmetric factor) have been calculated for each solute on each column. To find the similarities and dissimilarities among the columns and solutes principal component analysis (PCA) has been applied the columns being the variables and the retention parameters of solutes being the observations. The dimensionality of the matrices of PC loadings and components has been reduced by varimax rotation, nonlinear mapping technique and cluster analysis. Calculations were also carried out on the absolute values of the matrix elements. In order to assess the effect of PCA on the scattering of data nonlinear mapping and cluster analysis have been performed on both the original data matrix and its transposed form. The relationships between the results of data reducing methods have been elucidated by calculating linear correlations between the corresponding coordinates of the two dimensional maps.