Prediction Of LncRNA-disease Association Based On Graph Convolution Recommendation Algorithm

Long non-coding RNAs(lncRNAs)are a class of RNAs longer than 200 nucleotides that are not involved in coding proteins.A large number of studies have shown that lncRNA play important regulatory roles in various biological processes,moreover,their mutations and dysregulation are closely related to the occurrence and development of human diseases.Identifying the key roles of lncRNA in disease-related biological processes can accelerate the discovery of disease pathogenesis at the molecular level,which is helpful for clinical disease diagnosis and personalized treatment.However,traditional biology is very timeconsuming and costly in identifying lncRNA-disease associations.Based on the association information in existing databases and the help of effective computational methods,candidate lncRNA related to diseases can be quickly found to provide guidance for biological experiments.However,the existing lncRNA-disease association prediction methods are limited by the sparse known association data and the lack of negative samples.Generally,the prediction accuracy is not high,and it is difficult to apply to isolated disease-related lncRNA prediction.In order to solve the above problems,based on graph convolution recommendation algorithm(LGC),this article proposed two lncRNA-disease association prediction models.The specific research work is as follows:(1)This paper proposed a model called LGCLDA to predict lncRNA-disease association based on LGC.LGCLDA firstly constructed disease features by combining different biological information,it integrated lncRNA-disease,disease-gene and disease-mi RNA association data into disease node features.Then,the embedded information of diseases and lncRNAs was extracted by neighborhood aggregation,and the deeper topological information was obtained by three-layer embedding extraction.Subsequently,layer combination was performed to combine the initial embedded representation with the embedded information of each layer to obtain the final embedded representation of disease and lncRNA.Finally,the association between lncRNA and disease was obtained by means of inner product.LGCLDA does not require negative samples,it can predict isolated disease-related lncRNAs.The five-fold cross validation method was used to evaluate the model performance,and the experimental results showed that LGCLDA was superior to other comparison models.In addition,the case studies of lung cancer,colon cancer and gastric cancer further verified the validity of the model.(2)In order to further alleviate the problem of sparse data on known lncRNAdisease association,this paper proposed a lncRNA-disease association prediction model called BRLDA based on bidirectional recommendation pre-filling.In BRLDA,lncRNA node features were constructed based on lncRNA-disease,lncRNA-gene and lncRNA-mi RNA association data.LGC was used to predict lncRNA-related diseases,and the top K diseases were selected to prefill the known lncRNA-disease association data.The pre-populated data set was combined with the information of disease-related genes and mi RNAs to reconstruct disease characteristics,and LGC was used again to extract disease embedding information to predict disease-related lncRNAs.BRLDA and LGCLDA have the same advantages,not requiring negative samples,performing lncRNA association prediction for isolated diseases,and alleviating the problem of data sparseness.Compared with other prediction models,the experimental results of five-fold cross validation showed that BRLDA has better prediction capability.In addition,the case analysis of three diseases also showed that the model can achieve good prediction effect.