Research On Virtual Screening Of COX-2 Inhibitors Based On Machine Learning Methods

Machine learning can find specific links from the massive data and extract the trend and mode hidden in the data,so that we can understand the expression of data in the real world more intuitively.Machine learning technology can save a large amount of time in the phase of drug discovery.COX-2 inhibitor is a kind of antipyretic and analgesic drug,which is widely used in the anti-inflammatory and analgesic treatment of rheumatoid arthritis and osteoarthritis.COX-2 is an inducible enzyme,which is highly expressed in inflammatory cells and tumor cells and enhances the inflammatory response.The traditional Non-Steroidal Anti-Inflammatory Drug(NSAIDs)simultaneously inhibits COX-1(the isozyme of COX-2,protects the gastrointestinal tract and regulates platelet synthesis)and COX-2 activity,which can cause gastrointestinal damage to the chronic users.Therefore,it is necessary to find a specific inhibitor for inhibiting COX-2.In this paper,the machine learning theory is applied to the virtual screening of COX-2 inhibitors,the machine learning algorithm can identify the characteristics of COX-2 inhibitor sample data,therefore,a virtual screening model based on machine learning can be established to identify potential COX-2 inhibitors in the compound database.In this paper,a virtual screening model based on molecular descriptors and machine learning algorithms is constructed.The compounds were digitally processed with molecular descriptors,F-score algorithm is used for feature selection of molecular descriptor sets,reducing the dimension of sample data.SOM neural network algorithm is used for cluster the sample data which is divided into training set and test set,that guarantee the training set representative of the sample data.The random forest algorithm and SVM algorithm are used to establish the classification model of virtual screening,then these two models are compared and discussed.The results show that the combination of F-score algorithm can effectively improve the prediction accuracy of virtual screening based on random forest algorithm,compared with the classification model based on SVM algorithm,the overall prediction accuracy of classification model of random forest is 87.88%,better than the SVM model which is 81.52%.