Systematic comparison of statistical analyses of electronic and vibrational circular dichroism for secondary structure prediction of selected proteins.

The electronic (ultraviolet) circular dichroism (UVCD) and vibrational circular dichroism (VCD) of 20 proteins are systematically compared as to their relationship to the secondary structures of these proteins. The UVCD spectra are statistically treated by use of the same factor analysis methods used previously for VCD. The UVCD spectra can be reproduced as linear combinations of five subspectra. The first subspectrum reflected the expected alpha-helical UVCD shape, particularly at longer wavelengths, while the higher order ones had less obvious similarity to standard bandshapes. Cluster analysis on the UVCD factor analysis coefficients reflected the clustering on the basis of the fractional secondary structure parameters (from X-ray) but was less clear than VCD. Qualitative complementarity of protein VCD and UVCD spectra was demonstrated by combined cluster analysis of their respective factor analysis coefficients. Quantitative relationships between spectral coefficients and fractional secondary structure were determined by multiple regression analyses using only statistically important coefficients. These resulted in an ability to reproduce four of the structural parameters with errors for individual proteins comparable to the VCD result. In UVCD, the standard deviations of the regression fit for beta-sheet were worse and for the undefined part of the structure were better than in VCD. Parallel analyses using the partial least-squares method showed UVCD in that case to have more error than VCD in reproducing the training set structural parameters. Comparison of the regression and partial least-squares methods illustrated limitations of total back-transformation of the UVCD spectra into structural parameters.