Study On SAR And QSAR Of Insect Phenoloxidase Inhibitors

Phenoloxidase(PO,EC.1.14.18.1),also known as Tyrosinase,plays an important physiological role in the normal development of insects and is an important target for discovering new insecticides.In this paper,SAR(or classification)models and quantitative structure—activity relationship(QSAR)models have been established to predict the bioactivity of insect phenoloxidase inhibitors using various machine learning algorithm,and we used external test data set to evaluate the performance of the models.The main work of this paper as follows:First is the classification study,we established 3 kinds 9 models to distinguish highly and weakly active insect phenoloxidase inhibitors based on different data sets and machine learning methods.The first kind of models based on 65 insect phenoloxidase inhibitors from literature review,and contained 2 models using Support Vector Machine(SVM)and Decision Tree(DT)methods respectively.Since data of insect phenoloxidase inhibitors is limited,the SAR models may not have a good generalization effect,so we introduced 948 mushroom tyrosinase inhibitors which have similar structure and character with insect phenoloxidase inhibitors to expand the original data set,and using Support Vector Machine(SVM),Random Forest(RF)and Backpropagation Neural Network(BP)methods to establish 6 models.Among them,support vector machine model performed the best with an accuracy of 89.15%for the training set,an accuracy of 83.17%for the test set and an MCC value of 0.65 for the test set.However,for the external test set contained 13 newly reported highly active molecules,the accuracy of this model was less than 10%.Therefore,the research established the third kind of model using TrAdaBoost algorithm of transfer learning method to improve the model performance.The prediction accuracy of the external test set increased significantly,up to more than 60%when phenoloxidase inhibitor dataset took 10%-30%,and indicating that the migration learning algorithm was very effective for model improvement.Second part is the QSAR study.Based on 65 insect phenoloxidase inhibitors,using the different combination of Global molecular descriptors,2D Autocorrelation descriptors and 3D Autocorrelation descriptors,we established 6 QSAR models using Support Vector Machine(SVM)and Multilayer Perceptron(MLP).Among them,the models using all of the three kinds of descriptors have the best performance,where the multivariate linear regression model's correlation coefficient(r)of training set was 0.97,and r of test set was 0.93;the support vector machine model's r of training set was 0.96 and was r of test set was 0.91.Besides,this research also used these models to predict the bioactivity of 70 manchurian walnut components which have insect phenoloxidase inhibitors' potential.To sum up,the classification models and QSAR models could help us to predict the activity of a new insect phenoloxidase inhibitor,and the research provided a strong support for the research and development of new kinds of botanical insecticides.