Application Of Network Learning Algorithm In COVID-19 Drugs Screening

New coronavirus pneumonia,which is infected by new coronavirus,threatens the health of all mankind and causes huge economic losses to the world.Although new coronavirus vaccine has been vaccinated in many countries and regions,it still cannot avoid the risk of infection.Therefore,it is still urgent to screen possible therapeutic drugs.Based on the complete genomics sequence of the virus and the chemical structure of the drug,based on the network learning algorithm,two models are proposed to predict the relationship between the virus and the drug so as to screen the possible therapeutic agents for new coronavirus pneumonia.The main researches are listed as follows:Based on the virus similarity network,drug similarity network and virus-drug association network,a virus-drug association prediction model based on KATZ method VDA-KATZ is proposed to screen for the treatment of COVID-19 pneumonia.Firstly,the virus-drug association data set is sorted out;Then,the Gaussian kernel function is used to solve the Gaussian association profile kernel similarity of virus and drug.The Gaussian association profile kernel similarity of virus and its complete genomics sequence similarity,the Gaussian association profile kernel similarity of drug and its chemical structure similarity are linearly combined to obtain the final virus and drug similarity;Secondly,build heterogeneous network and integrate adjacency matrix to provide data basis for model calculation;Finally,the candidate drugs for the treatment of novel coronavirus pneumonia are screened by VDA-KATZ model,and the predicted drugs are docked with S protein and human ACE2 by molecular docking technology to further verify the accuracy of the model prediction.The experimental results show that compared with the firstclass models in the three association prediction directions,this model obtains the highest AUC and accuracy;Remdesivir and oseltamivir are the top two drugs predicted by the model,and their molecular binding energy is relatively high;In addition,elvitegravir and zidovudine have high binding energy with S protein and ACE2.Based on the idea of matrix completion,a virus-drug association prediction model VDA-GBNNR based on bounded nuclear norm regularization is designed to identify drugs for COVID-19 treatment.Experimental results show that compared with the four representative methods,VDA-GBNNR achieves the highest AUC values on dataset 2and dataset 3.Through the case analysis of drugs,it can be seen that most of the top 10 small molecule drugs predicted by the model have been studied and discussed by researchers in the recently published journals.Drugs such as remdesivir,favipiravir and niclosamide may have antinovel coronavirus effects by docking with the crystal structure of the RBD of S protein /ACE2 through molecular docking technology.