Application Study On Several Kinds Of Data Mining Methods In QSAR Prediction Of Drug And Toxicology

Quantitative structure-activity relationships(QSAR)is a set of methods based on the calculation method,such as statistical analysis,data mining and artificial intelligence,etc.,in order to study the relationship between the chemical physical properties or activity of all kinds of material and its structure.QSAR can excavate the invisible essence rule between structure and activity,or may dig out of the inherent rule of the experimental data directly.QSAR research in computer science,chemistry,materials science,biology,biotic and medicine has the vital significance.In this thesis,two aspects of QSAR model and molecular structure selection are discussed based on the problems of QSAR study in drugs and poisons.The main contents and results are as follows:1.PSO BP ANN model based on particle swarm optimization(PSO)and back propagation artificial neural network(BP ANN)was proposed and applied to QSAR prediction of the acidity coefficient(pKa)of a number of neutral and alkaline drugs.The results show that the PSO BP ANN model has a better prediction performance than other similar models,it is a better precision and correlation,and is one of good methods of QSAR study.2.PSO KHM RBF ANN model based on PSO,K-harmonic clustering(KHM)algorithm and Radial basis function artificial neural network(RBF ANN)was proposed and applied to QSAR prediction of gas chromatographic retention time of volatile organic compounds.The experimental results show that the PSO KHM RBF ANN model has better prediction precision,less training and testing error,higher correlation,and can provide new ideas for QSAR modeling.3.BBPSO KHM RBF ANN model based on back-bone particle swarm(BBPSO),KHM and RBF ANN was proposed and applied to toxicity prediction of a number of alcohols organic small molecule compounds on the frog tadpole.Research also shows that BBPSO KHM RBF ANN model has better performance and can provide reference for QSAR modeling.4.Due to the good capability of global optimization,genetic algorithm(GA)is used to select the molecular descriptors.Results show that the GA algorithm has better performance of descriptors selection,is a reliable method to select molecular descriptors.