Research On Prediction Methods Of Microbe-Host Association

More and more evidence shows that microorganisms that are parasitic in the human body and on the surface of the skin play an important role in the development of many human diseases.In recent years,global infectious diseases caused by microorganisms such as influenza A virus(H1N1),and new coronavirus pneumonia(COVID-19)have endangered human health and have a huge impact on social life.By predicting the association between microorganisms and hosts,researchers can find potential intermediate hosts and then control them in a targeted manner,so as to achieve the purpose of human intervention in the transmission of microorganisms.However,it is undoubtedly time-consuming and labor-intensive to use biological experiments to verify the association between each possible microbe-host pair.Therefore,if a faster prediction model can be used in advance to identify those microorganism-host pairs that are more likely to have an association,then biological experiments can be carried out on these microorganism-host pairs that are more likely to have an association.At this stage,although the types of association relationship prediction algorithms are relatively abundant,there are not many algorithms that specifically address the microbial-host association relationship prediction problem,and most of the existing microorganism-host association relationship prediction algorithms are only applicable to specific species of microbial protein-The problem of predicting the relationship between host proteins is not outstanding in the prediction of the relationship between microorganisms and hosts in multiple species.To solve the above problems,this paper proposes a new microbial-host association relationship prediction algorithm ISBIKATZ based on heterogeneous network,and balanced measurement model.First,we separately integrated the Gaussian association profile kernel similarity and cosine similarity of the microbes and the host's Gaussian association profile kernel similarity and cosine similarity to generate two new fusion similarity matrices and based on these two new The fusion similarity matrix constructs the microbial similarity network and the host similarity network respectively.Next,we obtained the microbial-host recommendation score matrix and the host-microbe recommendation score matrix by using a collaborative filtering algorithm based on fusion similarity and spectral clustering on the microbehost association relationship network and used these two matrices as adjacency matrix constructs a two-way weight network of the microbial-host association relationship.Then,to use more information,we formed a heterogeneous network by integrating the microbial similarity network,the host similarity network,and the microbial-host association two-way weight network.Finally,the final microbial-host association score matrix is obtained on the heterogeneous network with the help of the improved KATZ measurement algorithm introduced by the self-balancing factor.To evaluate the prediction performance of the ISBIKATZ algorithm,we selected five comparison methods using commonly used performance evaluation indicators(The AUC value and the AUPR value)are compared horizontally.The results show that the prediction performance of the ISBIKATZ algorithm is the best on the above three data sets.At the same time,to further verify the credibility of ISBIKATZ's prediction results,we conducted a biological verification of the top 20 potential microbial-host associations predicted by searching public literature and finally found that 15 of them were supported by biologically experimental verification.In summary,the ISBIKATZ algorithm is an ideal method for predicting the association between microorganisms and hosts.Finally,based on the previously collected microbe-host association data,we built a microbial-host association database and online prediction platform MHAP.Its purpose is to provide convenience to related researchers.In addition to visually displaying microbial-host association data in the form of graph view and table view,MHAP also provides search,submission,download and online prediction functions.