The Application Of Support Vector Machine In Research Of Quatitatives Structure-Activity Relationships Of Organic Compouds

The research of quantitative structure-activity relationships(QSAR)of organic compouds and drugs is of great importance in drug design.The aim of QSAR study is to describe the relationship between the activity and the molecular structures using mathematics and statistical methods.It provides an important way in drug design and improves the efficiency of developing a new drug.Support Vector Machine(SVM)is a powerful learning algorithm of artificial intelligence which shows high efficient abilities for classification and prediction of organic compouds and drugs.In the first part of this thesis,several quantitative structure-property relationships(QSPR)models between the relationships of the alkyl benzenes' structures and their heat capacity(C_p)and the standard enthalpy(â–³H_f~0)and burning heat(H_c)were built.A simple set of six numerical codes was used to represent each structure of the alkylbenzene which was derived from its molecular formula.We used multiple linear regression(MLR),multiple nonlinear regression(NMLR)and support vector machine(SVM)regression to establish the QSPR models.All models show good prediction abilities and the correlation coefficient R of all models are above 0.97.In addition,for three C_p models,the root mean square(RMS)errors are less than 12,the mean abstract error(MAE)are less than 8;for threeâ–³H_f~0 models,the RMS errors are less than 9,the MAE are less than 8;for three H_c models,the RMS errors are less than 150,the MAE are less than 130.Among all the above models, support vector machine are better than the other two ways.In the second part of this thesis,The classification and quantitative structure-activity relationships models of the aqueous solubility of 1293 organic compounds were built.The solubility is one of the important ADME properties of organic compounds and it influences the absorption and the biological activity of the compouds.It is important for filtering drugs by classifying the compounds according to solubility using support vector machine.The 1293 molecules were represented by 18 topological descriptors.First,the classification model was developed by SVM.The classification result achieved 92.2%for the test set.In addition,a quantitative model for the prediction of solubility of 1293 compounds was generated by SVM.For the test set,a square of correlation coefficient of 0.95 and a standard deviation of 0.50 were achieved.All models built by support vector machines show good prediction ability. In summary,some models for predicting the properties of organic compouds were built using SVM.It can give helpful information for further study of quatitatives structure-activity relationships of organic compouds and drugs.