Analysis of Deoxyribonucleotides with Principal Component and Partial Least-Squares Regression of UV Spectra after Fourier Preprocessing

Publisher Website
Google Scholar
Abstract
Principal component (PCR) and partial least-squares (PLS) regression with Fourier preprocessing of ultraviolet spectra have been evaluated for the analysis of mixtures of the four major nucleotides of DNA. Results have been compared with those obtained without Fourier preprocessing and with those obtained with a straightforward P-matrix approach with optimum wavelength selection. Calibrations were developed and tested with the use of sixteen standard and fourteen unknown mixtures. The standards and nine of the unknowns followed the typical base equivalences found in DNA. For these nine unknowns, the best average percent error in the Fourier domain was 0.61%, as compared to 0.65% in the spectral domain. The remaining five unknowns, in which the nucleotide concentrations were independent of each other, provided a particularly difficult test of each method's ability to distinguish the components. For these unknowns, the best errors in the Fourier and spectral domains were 1.20 and 1.85%, respectively. For assessment of the effect of protein, the analysis was repeated with varying amounts of albumin up to 36% by weight added to the mixtures. All of the full spectral methods (PCR and PLS in the spectral and Fourier domains) were able to compensate for the spectral interference of the added protein and produced results as good as those without the protein. For nucleotide mixtures, ultraviolet spectroscopy represents a rapid, economical, sensitive, and nondestructive alternative to existing techniques.