Research On Prediction Method Of Drug-Target Interaction Based On Network Model

The rapid identification of drug-target interactions can play a key role in the drug development process,and it can shed light on potential therapeutic properties.However,due to the experimental determination of drug-target interactions is expensive and time-consuming,there are still a large number of unknown drug-target interactions waiting to be verified.Therefore,many researchers have begun to use intelligent algorithms to solve the problem of predicting drug-target interactions in recent years.Although these methods can effectively predict drug-target interactions to a certain extent,they ignore the two problems that the link generation mechanism in the drugtarget network and the related auxiliary data of existing drugs or targets are also missing a lot.To solve the above-mentioned problems,this paper proposes two research methods for mining potential associations between drugs and targets.The main tasks are as follows:(1)A drug-target interactions prediction method,named NFSPDTI,is proposed based on multi-source data nonlinear fusion and structural perturbation.This method takes into account that the existing methods ignore the link generation mechanism in the drug-target network,NFSPDTI makes full use of the link generation mechanism characteristics of the drug-target network based on the idea of structural perturbation.First,the method separately calculates multiple drug similarity and multiple target similarity,and then used the nonlinear similarity diffusion algorithm to integrate different drug similarity information and different target similarity information to construct a more reliable drug and target similarity network,respectively.Next,a bilayer network is designed using the obtained drug and target similarity diffusion network and known DTI network.Finally,the idea of link generation mechanism is used to predict the potential association between the drug and the target.Experimental results show that compared with existing algorithms,NSFPDTI can more accurately identify potential DTIs,and the prediction performance is more stable under different experimental settings.(2)A drug-target interactions prediction method,named MCLNR,is proposed based on matrix completion and linear neighbor representation.This method introduces the idea of matrix completion technology to complete the imperfect similarity data before the prediction,which solves the problem that the traditional DTI prediction method ignores a large number of missing existing drugs or target-related auxiliary data.First,based on the side effects of the drug,the PPI of the target,and the known drug-target interaction data,two similarities of drug and two similarities of target were calculated respectively.Then,the idea of matrix completion technology was introduced to complement the imperfect similarity data to enhance their information abundance,thereby constructing a similarity network with greater information volume.Next,a weighted combination strategy was proposed to integrate drug similarity and target similarity to obtain comprehensive drug similarity and target similarity.Afterward,a heterogeneous network was constructed by using the obtained drug comprehensive similarity,target comprehensive similarity and known DTI network.Finally,the idea of linear neighbor representation learning is used to predict potential DTIs.Experimental results show that MCLNR can effectively predict potential DTIs,and most of the predicted new associations have experimental evidence.