Variable Optimization Is Used For Quantitative Structure-activity And Spectral Modeling

With the rapid development of computers technics, which assisted predict more and more attention in Drug Design, especially Quantitative Structure-Activity Relationship(QSAR) and Molecular Docking. QSAR is mainly through the research of the structure of a series of compounds and their corresponding biological properties and the relationship between the physical and chemical properties. While the molecular Docking based on "the lock- key principle", simulating the interaction between two or more molecules recognition process. Here in, we studied the diaryl pyrimidine analogues(DAPY) for the inhibition of wild type HIV-1, Cytochrome P450 1A2 for 52 compounds(including 19 naphthalene and its derivatives, six quinoline/isoquinoline derivatives, 15 lactone, 3 biphenyl and 9 other compounds) of inhibition effect and the data was measured of based on near infrared spectrum for corn protein, fat, starch and moisture content has been forecasted. The established models has a good stability, predictive, and find out the main structure factors affecting inhibitors which provide ideas for the synthesis of new inhibitors.This thesis includes four parts, the first part we briefly introduced the research present situation and development of computer, the development of the Quantitative Structure-Activity relationship, research steps, the principle of four kinds of modeling method and Molecular Docking technology.The second chapter we studied the uses of the stepwise multiple linear regression(stepwise-MLR) method selected seven descriptors, Based on the seven descriptors and half inhibitory concentration of 32 DAPYs compounds established PLS, SIMPLS, SVM and PSO- SVM model, predicted the 32 DAPYs compounds of inhibitory concentration. By comparing statistical parameters and graphics of four models that SVM and SIMPLS models have good stability and prediction performance.The third chapter is based on the molecular structure of the 52 compounds and inhibitory activity of cytochrome P450 1A2 inhibitor by the four methods of PLS, SIMPLS, SVM and PSO- SVM were established QSAR model. The SVM and PSO- SVM model of correlation coefficient in the first category is higher than the second and the third class, the bottom of the root mean square error of 52 compounds for inhibition of CYP 1A2 has better inhibition. By analyzing the models found that SIMPLS in the four methods has more reliable prediction performance.The last chapter, the samples of protein, starch, moisture, oil content in the training data set and the corresponding wavelength of near infrared spectral characteristics, using PLS combined with Wavelet method was established model of corn grain protein, starch, moisture, oil content. It showed that the PLS combined with wavelet transform of the model prediction effect is better, the established prediction model is stable, has high prediction precision and the reliable prediction ability, has reached the expected effect.