Research On Predicting Disease-related MicroRNA Based On Topology Information

Research shows that the occurrence of disease is related to the abnormal regulation of miRNAs(microRNAs),thus to discover disease-related miRNAs can contribute to study the pathogenesis and treatment of diseases.However,it is expensive and time-consuming to obtain accurate miRNA-disease associations by experimental verification.Therefore,computational-based methods for predicting miRNA-disease associations have become research hotspot and are of great significance.Specifically,by exploiting the known miRNA-disease associations,the possibilities of potential associations between miRNAs and diseases are calculated and sorted according to the probability score,so as to recommend candidate miRNAs with higher scores to the researchers.Based on the biological network and the internal topological information of the network,this paper studies the disease-miRNA link prediction algorithm.The main work is as follows:(1)Construct disease-miRNA bilevel network,and a prediction method based on network representation learning is proposed by network embedding to combine the weight and association information in network.The validity and superiority of the proposed method are verified by the analysis of prediction results and the comparison of classical algorithms.(2)Propose two improvements from different perspectives base on the prediction method of representation learning on network.Firstly,improve network structure by adding gene nodes,thus expanding network topology information.Secondly,change network coding mode by DeepWalk algorithm,thus avoiding constructing similarity networks.The result and algorithm comparison show that the improved algorithm is effective and can greatly improve the effect.(3)In order to utilize network topology information,a disease-related microRNAs prediction method based on machine learning model is proposed.The method combines external biological information and internal topological structure information by constructing features from similarity matrix and topological matrix.The experimental results show that the proposed algorithm can effectively predict disease-miRNA associations,and is significantly better than the contrast method.The top 30 candidate microRNAs recommended in case analysis can be basically confirmed by the databases.