Cheng Yiyu

Fractal fingerprinting of chromatographic profiles based on wavelet analysis and its application to characterize the quality grade of medicinal herbs.

Extracting chemical fingerprints is an important step for representing and interpreting chromatographic data. In this paper, the chromatographic profile is decomposed into components at different resolution levels using wavelet analysis, then the fractal dimensions of these components are computed as the chemical fingerprints. The chromatographic fingerprint is characterized by the vector composed of these chemical fingerprints, which can represent the chemical patterns of different categories of complex samples. Computer simulations reveal that the fractal fingerprints are more stable than the original chromatographic profile data with respect to variations of peak retention time. To demonstrate the validity of this method, the evaluation of the quality of the medicinal herb Angelica sinensis (Oliv.) diels is investigated. Principal component analysis of the fractal fingerprints indicates that samples belonging to the same quality grade are clustered together, while those belonging to different quality grades are separated. Using these fractal fingerprints taken from the chromatographic scans as inputs for an artificial neural network (ANN). The quality grades of two sets of the herbs were verified by cross-validation, indicating that 96.7% of the herbs are correctly identified with respect to their quality grades evaluated by experienced experts, and 100.0% of the herbs are correctly identified with respect to their quality grades determined by pharmacodynamical evaluation.

Gas Chromatography-Mass Spectrometry Based on Metabonomics Study of Carbon Tetrachloride-Induced Acute Liver Injury in Mice

Abstract A gas chromatography-mass spectrometry based on methodology was developed to investigate the distinctions of metabolic profiles in liver tissues and plasma of mice with different time post-dose after CCl 4 treatment. NIST mass spectral library was used for the rapid identification of endogenous metabolites, and the principal component analysis was introduced for the pattern recognition of mice metabonome. The pattern recognition result presented clear changing tendencies of metabolite profiles at various time postdose, both in liver tissues and plasma. The elevated levels of malate and several fatty acids in liver tissues as well as the increased concentrations of some amino acids and citrate in plasma were detected as metabolic effects of CCl 4 induced liver injury. Changing tendencies of the metabolite profiles and several metabolites were conformed to the histopathological assessment of the liver tissues. The results suggested that the GC-MS based method can provide some advantages in metabolite profiling and compound identification, and therefore can be employed as a potential powerful approach in metabonomics research as a complement of NMR.

An improved genetic algorithm for training layered feedforward neural networks

The new genetic algorithm for training layered feedforward neural networks proposed here uses a mutation operator for performing the search behaviors of local optimization. Combining the random restart method with the local search technique, the algorithm can converge asymptocally, to the optimal solution. Test with a practical example showed that the improved genetic algorthm is more efficient than the conventional genetic algorithm.

A new computing multivariate spectral analysis method based on wavelet transform

This paper proposes a new algorithm for multivariate calibration named Principal Component Regression Based on Wavelet (PCRW) which combines wavelet decomposition technique with the factor analysis method for establishing a duplicate denoising mechanism. A practical example in spectral analysis of a typical multicomponent pharmaceutical system was used to verify the effectiveness of the algorithm. It was shown that PCRW produced fewer prediction errors than those obtained by using PCR.