Computer Aided Study On The Structure-Activity Relationship Of Leukotriene A4 Hydrolase Inhibitors

Leukotriene A4 hydrolase(LTA4H)can promote the conversion of leukotriene A4(LTA4)into leukotriene B4(LTB4),which is related to inflammatory disease such as asthma,inflammatory bowel disease and stroke.At the same time,LTA4H also exists in cancer cells such as skin cancer and colon cancer.Therefore,LTA4H is an important anti-inflammatory drug target.To study the leukotriene A4 hydrolase inhibitors,we used computational chemistry and some machine learning algorithms to build the classification models and quantitative regression models,and analyzed the structure and features of inhibitors,and virtual screened hits.The main research contents of this paper are as follows:(1)Build classification models for LTA4H inhibitors.We collected 463 LTA4H inhibitors and divided them into highy active inhibitors(300)and weakly active inhibitors(163)under the threshold(IC50=0.6 μM).Then we used self-organizing neural network division and random division to divide data to training set and test set.We used CORINA,MACCS fingerprints and ECFP4 fingerprints to characterize the structure of the compound,and built 18 classification models by support vector machine(SVM),random forest(RF)and K nearest neighbor(KNN).Model 2A performed well in the training set and the test set.The accuracy of the training set and the test set were 87.38%and 88.96%,respectively.The MCC value was 0.74 on the test set.Then we clustered 463 LTA4H inhibitors into six subsets by K-Means algorithm,and found that the scaffold structures of the highly active inhibitor of LTA4H were diaryl ether and benzothiazole.In addition,hydrogen bonds and π atoms electronegativity could explain the interaction between LTA4H and the ligand.(2)Build quantitative regression models for LTA4H inhibitors.We collected 172 LTA4H inhibitors whose IC50 values were determined by ELISA,and selected CORINA and RDkit molecular descriptors to characterize the compounds.We divided data to the training set and the test set three times randomly,and built 12 quantitative regression models by multiple linear regression(MLR)and support vector machine(SVM).Model 5A performed best.The coefficients of determination(R2)of the training set and the test set were 0.81 and 0.79,and the absolute error(MAE)were 0.32 and 0.35,and the mean square error(MSE)were 0.16 and 0.20,respectively.By analyzing the descriptors,we found that hydrogen bonds,electronegativity of π atoms and molecular polarity had important effects on compounds.(3)Use two best classification models(Model 1C and Model 2A)and the best quantitative regression model(Model 5 A)to screen compounds as method one,and use shape similarity and electrostatic similarity to screen compounds as method two.By screening the Specs database containg 210596 compounds,we obtained ten compounds as hits by two methods.Among them,the predicted pIC50 value of compound No.2 was 7.200,and its scaffold structure was benzothiazole,which was the same as the main scaffold structure of the highactivity inhibitors;the predicted Tanimoto value of compound No.7 was 2.010,and its scaffold structure was biphenyl.In this thesis,we built the classification and quantitative regression models of LTA4H inhibitors,analyzed their structure and obtained the scaffolds related to the highly active inhibitors.Through virtual screening of the compound database,we obtained some hits.These methods provided a means to study the properties of LTA4H inhibitors.