Research On The Prediction Method Of Microbe-drug Association Based On Similarity Information

The emergence of microbial resistance has led to a steady de-cline in the therapeutic efficacy of antibiotics,and the evolution of antimi-crobial drug resistance represented by antibiotics is now in accelerated for-mation.Identifying the relationship between microorganisms and drugs can help us deeply understand the interaction mechanism between drugs and mi-croorganisms,which is helpful for the screening of candidate compounds and accelerating the discovery of new drugs.It is expected to be one of the ways to solve the problem of drug resistance.Traditional biochemical experiments to verify the correlation are time-consuming and costly,while computational methods for microbe-drug association prediction can rely on the powerful computing power of computers to make predictions,provide reference for drug developers,reduce the number of biological experiments,improve the efficiency of drug development,and save human and material resources.In this paper,we propose two different microbe-drug association prediction methods based on biological data related to drugs and microor-ganisms,combined with machine learning related algorithms.The main research contents and innovation points are as follows:(1)The current association matrices composed of microbe-drug asso-ciations are sparse matrices,and the sparsity of association matrices has a certain influence on microbe-drug association prediction.To address this issue that the existing microbe-drug association prediction methods do not consider reducing the influence on model prediction performance by reduc-ing the sparsity of association matrices,this paper proposes a pre-completion-based label propagation algorithm PLPMDA for microbe-drug association prediction,which reduces the impact of its sparsity on prediction by pre-completing the original microbe-drug association matrix.Specifically,the sparsity of the association matrix is reduced by pre-completing the origi-nal microbe-drug association matrix using the weighted known K nearest neighbors(WKNKN)method which uses the similarity information of mi-crobes and drugs,and then the similarity of microbes and drugs is calculated using the linear neighbor similarity measure for the pre-completed associ-ation matrix,and finally the label propagation algorithm is used to predict the microbe-drug association.The results of the ablation experiments show that the measures such as the pre-completion treatment of association ma-trix and the linear neighbor similarity measure method used in this paper can effectively improve the prediction performance of the model.The re-sults of de novo experiments and cross-validation experiments show that PLPMDA performs much better than other comparative methods.The case studies of microbe and drug prediction also show that PLPMDA method can predict potential microbe-drug association relationships well.(2)In bioinformatics,networks are widely used to represent biomed-ical entities(as nodes)and their relationships(as edges),where network nodes contain rich textual information.To solve this problem that most methods of network representation learning only utilize network-related structural information and ignore network node information,or simply stitch the learned network feature representations with related node information.In this paper,we propose a network representation learning-based microbe-drug association prediction method TADWMDA,which can effectively com-bine network structure information and network node feature information.Specifically,the similarity information of microbes and drugs is used as the node text information of the microbe-drug network,and the heterogeneous network constructed based on the similarity of microbes and drugs and the microbe-drug association relationship,TADWMDA method learns to ex-tract the microbe-drug association information and the similarity informa-tion of microbes and drug nodes from this network to obtain the microbe and drug node feature representations,put them into the support vector machine(SVM)classifier for microbe-drug association prediction.The results of cross-validation experiments show that the TADWMDA method effectively improves the prediction performance of the model by combining node-related information and network structure information,which makes it per-form better than other comparative methods.By visualizing and analyzing the node feature representations learned by this method,it can be seen that the node feature representations learned by the TADWMDA method can effectively fuse node similarity information as well as relevant association information.