Research On Association Prediction Of LncRNA-disease Based On Network Multi View

In the biological development process,with the continuous improvement of technology and experimental level,more and more studies have shown that lncRNA plays a key role in a large number of biological processes.However,the experimental verification of the role of lncRNA has certain drawbacks,such as time-consuming,labor-intensive and high cost.Therefore,we will study the inherent relevance of lncRNA to diseases from another perspective,namely the calculation method.In this article,two models based on hybrid algorithm and unbalanced two-way random walk and based on multi-view and multi-identifier learning are proposed to predict the association between lncRNA and disease.The content is as follows:(1)Based on hybrid algorithm and unbalanced bi-random walk(HAUBRW): The first step is to preprocess the data.According to the original matrix of lncRNA and disease,use the WKNKN algorithm to reconstruct the original association matrix,and then integrate them separately lncRNAs similarity(lncRNA functional similarity,lncRNA cosine similarity and lncRNA expression similarity)and disease similarity(disease semantic similarity and disease cosine similarity);the second step is to combine heat diffusion algorithm(Heat S)and probability diffusion algorithm(Prob S)combined hybrid algorithm will reallocate resources;the third part,use unbalanced double random walk(UBRW)to infer undiscovered lncRNAdisease associations.Seven classic and advanced models(BRWLDA,DSCMF,RWRlnc D,IDLDA,KATZ,Ping and Yang models)are compared with our method.The simulation results show that under the evaluation indicators of LOOCV and 5-fold CV,our method The AUC values are 0.8693 and 0.8617±0.0064,respectively.The results show that HAUBRW has better predictive performance.In addition,we have further studied the use of the method,HAUBRW can effectively predict the candidate lncRNA of breast cancer,osteosarcoma and cervical cancer.Therefore,our method may be a good choice in future biomedical research.(2)Based on multi-view multi-marker learning(MVML): The first step is to use known lncRNA and disease associations to construct multiple similarity matrices,namely lncRNA expression similarity,lncRNA functional similarity,lncRNA Gaussian nuclear similarity,Disease semantic similarity and disease Gaussian nuclear similarity.The second step is to learn lncRNA and disease network separately from the similarity of multiple views,and obtain the multi-view lncRNA similarity matrix and the multi-view disease similarity matrix.The third step is to continuously iteratively update the association between the lncRNA predicted from the two spaces and the disease based on the multi-signature learning.In addition,taking LOOCV as the evaluation index,in order to verify the performance of the MVML method,we also compared with five models(BRWLDA,HAUBRW,IDLDA,KAZLDA and Yang’s models).The results show that our AUC value is 0.8782,which is relative to other prediction performance The model is better.Let’s analyze the case of our method again.The MVML method can effectively predict the candidate lncRNAs for gastric cancer,ovarian cancer and colon cancer.