| MicroRNA(miRNA)plays crucial role in different biological processes and has been confirmed to be involved in the occurrence and development of diseases.Predicting the association between miRNA and human disease can facilitate the prevention and treatment of disease,and accelerate drug research.With the development of next-generation sequencing technology,available biological networks continue to emerge,providing valuable resources and challenges for the prediction of miRNA-disease associations.In recent years,deep learning and graph neural network have been widely used in various fields because of their fast and efficient modeling of practical problems.Therefore,this paper analyzes the strengths and weaknesses of existing miRNA-disease association prediction algorithms.Based on the multiple similarity views formed by miRNA and disease-related biological data,combined with deep learning algorithms,two miRNA-disease association prediction algorithms based on multi-view fusion and graph neural networks are proposed.The main work is summarized as follows:(1)Aiming at the problems that most existing miRNA-disease association prediction algorithms are difficult to effectively integrate multi-source data and ignore network topology information,a multi-view and multi-channel attention graph convolutional network algorithm MMGCN is proposed.The method uses graph convolutional neural networks to learn the features of miRNA and disease nodes on multiple similarity views,and utilizes the attention mechanism on multi-view features to adaptively learn the importance of different features.After combining the features with their corresponding importance,matrix completion is used to obtain potential miRNA-disease association score.Experiments show that MMGCN is superior to other comparison methods in all indicators.Meanwhile,ablation studies and parameter analysis demonstrate the necessity of each module of the MMGCN algorithm.Subsequently,experiments under different similarity views are carried out to prove the importance of multi-source information for miRNA-disease association prediction.Case studies further demonstrate the reliability of MMGCN in predicting novel miRNA-disease associations.(2)In order to further explore the potential association between miRNA and disease and improve the prediction accuracy of the model,a bipartite graph attention algorithm based on multi-view fusion(MVFGAT)is proposed to solve the problem that MMGCN cannot fully utilize miRNA-disease association network information.The model uses similarity view fusion algorithm to fuse multiple miRNA and disease similarity views information,and utilizes graph neural network to further extract node features.Then the bipartite graph attention encoder is used to extract node features of the bipartite graph composed of miRNA and disease,and the node features on the similarity network are spliced with the node features on the bipartite graph.Next,multilayer perceptron is used to merge the splicing features to obtain the final representation of miRNA and disease to infer the miRNA and disease association.Five-fold crossvalidation experiments validate the performance of MVFGAT,and subsequent experiments on ablation studies demonstrate the importance of each step of the model.Finally,the corresponding case studies illustrate that the model has strong potential as an auxiliary tool for biological experiment in practical application. |
