The Design Of Chemometrics Based Network Platform And Its Applications

The concept of chemometrics was introduced some 30 years ago. It is an interdisciplinary field (lying between measurement-oriented chemistry and applied statistics) that concerns the mining of information from chemical data with the help of mathematical and statistical means. Various kinds of chemometrics methods, namely principle component analysis, partial least square, chemical factor analysis, artificial neural network etc. have been accepted and employed widely. To help chemical engineers get more benefit from these methods, a lot of software, from commercial products MATLAB, SPASS to various chemometrics software packages and abundant web servers easily accessible on the Internet, have been invented and created. However, to meet the new standards and requirements of scientific computation, namely high availability, data security, large scale computation, user friendly interface etc., we introduced our cluster based chemometrics computing platform named RCPC. RCPC is comprised of 5 main parts: RCPC website was responsible for receiving user's task request and maintaining algorithms; control server is responsible for task scheduling and calculation servers'management, it is the core of RCPC; calculation servers are responsible for task computation, they contribute their computing power in cluster manner; algorithm library supports computing server with abundant algorithms and specifies a simple interface for algorithm integration; database server is used to save all the task-specific information. RCPC is tested in logical and physical way respectively. The logical test proves that task scheduler is in high efficiency and reasonably designed. The physical test confirms the robustness of RCPC system. The article introduced two important algorithms: one is support vector machine classification (SVMC) algorithm and the other is about 2-D structure illustrating of organic compounds. SVMC is a newly invented machine learning methods which are suitable for small data sets. Since the algorithm will eventually lead to a linearly constrained quadric programming problem, we implemented three solutions for this problem: grads decreasing algorithm, decomposition algorithm and SMO algorithm. We also give a way of probing for optimal parameter pair of SVMC algorithm. Structure illustration of organic compounds includes two steps: the first step will translate compound name into computer-readable data structure named GRAPH, the next step then draws the structure of that compound iteratively by reading GRAPH information. The article introduced three applications of RCPC system: Support Vector Machine Approach to Functional Group Prediction From Infrared Spectra; A FFT-based method for prediction of GPCR subfamilies; Study of infrared character of hydroxy compounds from FTIR spectra. The last part of this article introduced another network-oriented system of "GSP management system for medical business"the system is divided into 9 parts, covering every business aspects of buying, storing and selling. It can help the company reach the specification of GSP in everyday business.