UV-Visible Spectral Library Search with Mixtures

Abstract

A library mixture search method originally developed for infrared spectra has been successfully applied to UV-visible spectra. This novel approach for searching a spectral library performs a principal component analysis (PCA) on the entire library of spectra for pure compounds. The library spectra are represented by their PCA scores, and the concentrations (assumed to be unity) are regressed onto these scores. The scores for an unknown spectrum projected onto the PCA basis set are multiplied by the regression matrix to predict pseudo-concentrations or composition indices. After the first pass through the library, a subgroup of the top 20 hits (10% of the library) is selected and the PCR analysis is repeated on this set to improve the selection process. Spectra of each of the individual target components are adaptively filtered from the subgroup of library spectra and from the unknown spectrum prior to the repeat of the PCR analysis. The application of the adaptive filter greatly improves the success rate on hitting the second and third components by removing the first hit during each pass through the library. Computation times for training and applying the Mix-Match algorithm are greatly reduced by pre-processing with Fourier Transforms. A 200-compound library could be trained in 45 min and searched in 9 s; a 20-compound subgroup could be adaptively filtered and searched in 37 s. Both components in 12 two-component mixtures and one component in each of two two-component mixtures were correctly identified; the algorithm failed on both components in only one out of 15 two-component mixtures. All three components were correctly identified in one three-component mixture, and one component was correctly identified in another three-component mixture.