Research Of Protein-drug Interaction Prediction Based On Bipartite Network

Drug research and discovery is a long-term,high-risk,and costly systemic project.The average period of a new drug from research to marketing is 10-15 years,costing hundreds of millions of dollars.The use of calculation methods can effectively shorten the drug development cycle,so it has received extensive attention.With the rapid growth of high-throughput biological data,network-based protein-drug interaction prediction methods have gradually become a research hotspot.However,network-based methods have the problem of unclear biological action mechanisms,and the prediction performance needs to be further improved.This paper presents a network prediction model for the interaction between the residue-trimers of DNA binding proteins and drugs,and analyzes the intrinsic interaction mechanism between DNA binding proteins and drugs.In addition,a prediction method based on improved protein-drug link is proposed.The full text work consists of two parts:(1)The research on DNA-binding protein molecular fragment-drug interactions.Protein molecular fragments(trimers)are used to describe DNA-binding proteins.Based on the similarity of physicochemical properties,molecular fragments are clustered into different physical and chemical clusters.Then,a network is constructed based on the association relationship between drugs and clusters.The network can analyze the physicochemical properties of drug-trimer binding,and then explain the DNA-binding protein-drug binding mechanism.Based on the CN method in the link prediction model,10-fold cross-validation is used for prediction.The experimental results show that the network prediction model can effectively predict the interaction between protein molecular fragments and drugs,and achieve higher prediction accuracy.(2)The network prediction of protein-drug interactions.Based on the protein-drug interaction,a link prediction method(DLS)for the local structure of the network is proposed.The local structure mainly includes two parameters: node degree and path.At the same time,the theory of mass diffusion is introduced to define the protein-drug interaction score in the network,obtain the recommended protein of the drug according to the score,and find the potential target protein of the drug according to the recommended list to achieve the relocation of the drug.First,the network is established based on the known protein-drug interaction,and then the interactions between the protein and the drug in the network are predicted based on the DLS method.Compared with the traditional three methods,the DLS method achieves better performance in AUC,AUPR,accuracy,sensitivity,F1-score,and accuracy.And this method predicts candidate proteins for three drugs(captopril,desferrioxamine and losartan).