Dependent Component Analysis and its Applications for Estimation of Source Ultraviolet Spectral Profiles and Characterization of Processing Batch for Preparation of <i>R</i><i>adix s</i><i>cutellariae</i>

Abstract

Dependent component analysis (DCA), which is an extension of independent component analysis (ICA) for blind source separation (BSS) and requires no assumption on the distributions of the sources, was used to directly estimate source spectral profiles from ultraviolet spectra (UV) of mixtures. By simply assuming that the sources are dependent only through their variances and have temporal correlations, variance DCA was established. The efficiency of DCA for estimation of source UV spectral profiles was qualified by synthetic mixed UV data. It was shown that the estimation efficiency of DCA is better than that of FastICA when the sources are seriously overlapped. Then the DCA was used to directly estimate source spectral profiles from UV data that were measured at different steaming periods during the processing procedure batch for preparation of radix scutellariae. The estimated dependent components (DCs) and their variations of relative concentrations were used to characterize the processing batch. The results show that the estimated DCs are corresponding to sterol and flavonoid compounds, respectively. By inspection of the change trends of the estimated DCs, the endpoint of the processing batch was determined as 55 min, which is more accurate than that it should be located in 30~60 min by traditional sensory analysis. DCA provides an alternative approach for estimation of source spectral profiles from the overlapped spectral signals, and UV-DCA can be used as a novel way for characterization of the traditional Chinese medicines processing procedure.