Fundamental Study Of Informatics For Pharmaceutical Analysis & Metabonomics

With the arising of the "-ornics" methodologies, such as genomics, proteomics, and metabonomics, and the rapid development in modern analytical sciences, pharmaceutical analysis study has been broadened to a new field and the research objectives and analytical requirements has been correspondingly diversified. Meanwhile, the size and dimensions of datasets in pharmaceutical analysis sharply increase, which makes the conventional data analysis methods become inadequate. As a consequence, resolution, interpretation and decipherment of pharmaceutical data become the most critical problems, which instigate informatics for pharmaceutical analysis to be a frontier area in pharmaceutical analysis study. In this research, computational methodology for pharmaceutical analysis and metabonomics were investigated with insight. A number of novel informatics approaches for feature extraction, quantitative composition-activity relationship (QCAR), and chemical fingerprinting, integrating the intelligent computational methods and instrument analytical techniques, were developed according to the specific requirements of different fields of pharmaceutical analysis and metabonomics. The main results are stated as follows:1. Curse of dimensionality and dataset sparsity are two practical constraints in the analysis of complex pharmaceutical data, such as TCM data. To compensate the data complexity, two novel feature extraction methods were proposed: stepwise correlative component analysis (SCCA) and optimal classification analysis (OCA). These two methods were respectively applied to classify Angelica and Chuanxiong samples of different quality grades. The results showed that the classification capability of the features extracted by these two methods were better than that of features extracted by PCA, and the neural classifiers derived from these two methods have simple architecture and high classification accuracy. Additionally, a learning approach, which can adaptively construct the optimal architecture of multilayer feed forward neural network (MFNN), was employed for modeling QCAR of Traditional Chinese Medicines (TCM), and it was successfully applied to the prediction of ChuanXiong bioactivity.2. Chromatographic fingerprinting is a rapid developing research area in pharmaceutical analysis and quality control of TCM. The proposed novel methods in this study deal with anumber of critical problems occurring in the development of chromatographic fingerprinting.(D A novel concept, multiple chromatographic fingerprinting (MCF), which consists of more than one chromatographic fingerprints and represents all chemical composition characteristics of analyte, was proposed as a strategy for quality control of complex TCM. A new methodology on information-fusion-based computation was also proposed for evaluating the relationship of the multiple fingerprints. In the typical example, binary chromatographic fingerprinting of Compound Danshen Dropping Pill (CDDP) was developed for consistency assessment and frauds detection. The results indicated that the multiple chromatographic fingerprinting could show a prosperous prospect in quality control of complex TCM. The proposed MCF method was used by TASLY Pharmaceutical Co. as an approach to quality control of CDDP, and has been employed in fingerprinting research of Kuhuang Injiection and Shuangdan granule. Recently, 'fingerprinting and quality control technology of CDDP', of which MCF is the core technology, achieved an 'Award for Science and Technology of Tianjin (2nd class)'.(2) LC/MS fingerprinting of Shenmai Injection was constructed and applied to evaluate the quality. Ginsenosides and ophioponins were clearly represented in the LC/MS fingerprint, while the reported HPLC-UV fingerprint represents only the composition of ginsenosides. Further, the proposed LC/MS fingerprints were applied to identify the product manufacturers. All 21 samples were accurately identified based on their LC/MS fingerprints in conjunction with PCA.? A sort similarity determining method was developed in this study to address the problem of similarity evaluation of the chromatographic fingerprints of different classes. This method was applied to evaluate the quality of Sarcandra Glabra (Thunb) Nakai samples through their HPLC fingerprints. The results showed that the different parts of this plant, i.e. the aerial and the whole, were clearly classified, and chemical fluctuations of samples with the same sort were well represented.3. Metabonomics/Metabolomics is a novel "-omics" technology, initially proposed by Nicholson in 1999, and has been widely accepted as an important part of system biology. However, there has been little attention paid to the development of thorough and robust computationalmethodology for Metabonomics. Regarding this, cancer diagnosis was investigated as a typical example in the desired research field of computational methodology for metabonomics in this study. A computational method for analysis of the HPLC metabonomics fingerprints was proposed and used to construct a diagnosis model, distinguishing 25 patients with breast cancer from 19 healthy people by analyzing their urine samples. The predictive rate of the diagnosis model is up to 93.2% for test set. Furthermore, to enrich the information obtained in the instrumental analytical approach, Multi-source Metabonomics Fingerprinting (MMF), consisting of more than one metabonomics fingerprint which are acquired through several analytical instruments (or detectors), was proposed based on multiple classifiers combination techniques. The proposed method was applied to diagnosis of lung cancer using HPLC and CE metabonomics fingerprints, and the results showed that all samples (12 patients with lung cancer and 12 healthy people) were correctly classified. Due to the analogy among the corresponding computation methods in different fields of metabonomics, these researches provided some novel analysis tools for the application and dissemination of metabonomics.