Influence of predicted elution regions on the performance of methods for evolutionary factor analysis as applied to high-performance liquid chromatography

Methods for evolutionary factor analysis (EFA) (also called window factor analysis) as applied to simulations of two-way diode array high-performance liquid chromatography data are compared. Three indices of quality of reconstructions are proposed, namely correlation between true and estimated time profiles and spectral reconstructions, and weighted integration errors. Two approaches to EFA are described, namely EF0 involving using only composition 0 regions to determine factor rotations and EF1 involving using both composition 1 and composition 0 regions. Choice of regions where one compound elutes is found to have an overwhelming influence on quality of reconstructions; integration errors are far more sensitive indicators than correlation coefficients; EF0 is a form of exploratory independent modelling, where EF1 performs best when the data are better and is a form of dependent modelling.