Differentiation ofAurantii fructus immaturusandFructus poniciri trifoliatae immaturusby Flow-Injection with Ultraviolet Spectroscopic Detection and Proton Nuclear Magnetic Resonance Using Partial Least-Squares Discriminant Analysis

Abstract

Two simple fingerprinting methods, flow-injection coupled to ultraviolet spectroscopy and proton nuclear magnetic resonance, were used for discriminating between Aurantii fructus immaturus and Fructus poniciri trifoliatae immaturus. Both methods were combined with partial least-squares discriminant analysis. In the flow-injection method, four data representations were evaluated: total ultraviolet absorbance chromatograms, averaged ultraviolet spectra, absorbance at 193, 205, 225, and 283 nm, and absorbance at 225 and 283 nm. Prediction rates of 100% were achieved for all data representations by partial least-squares discriminant analysis using leave-one-sample-out cross-validation. The prediction rate for the proton nuclear magnetic resonance data by partial least-squares discriminant analysis with leave-one-sample-out cross-validation was also 100%. A new validation set of data was collected by flow-injection with ultraviolet spectroscopic detection two weeks later and predicted by partial least-squares discriminant analysis models constructed by the initial data representations with no parameter changes. The classification rates were 95% with the total ultraviolet absorbance chromatograms datasets and 100% with the other three datasets. Flow-injection with ultraviolet detection and proton nuclear magnetic resonance are simple, high throughput, and low-cost methods for discrimination studies.

Development of a Comprehensive Flavonoid Analysis Computational Tool for Ultrahigh-Performance Liquid Chromatography-Diode Array Detection-High-Resolution Accurate Mass-Mass Spectrometry Data

“…Therefore, the wavelength range of UV–vis spectrum needs to be optimized in this study. One straightforward way to achieve this is to build chemometric models for different wavelength ranges, evaluate the models by cross-validation methods such as leave-one-sample-out method , and bootstrapped Latin partition method, calculate the classification rates for the different wavelength ranges, and select the optimum range which gives the best classification rate. However, this calculation required hours to execute depending on which chemometric models and validation method were chosen and is not practical to use in the data processing program.…”