Research On Network Methods For MiRNA-disease Association Predictio

The increasing incidence and mortality rates of complex diseases pose a significant threat to human health.Micro RNAs(miRNAs),a class of non-coding RNAs approximately 22 nucleotides in length,play crucial roles in various physiological processes,including cell growth,differentiation,and signal transduction,making them closely associated with disease onset and progression.Therefore,predicting the association between miRNAs and diseases is of great significance for understanding disease mechanisms and facilitating early prevention,diagnosis,and treatment.Traditional biological methods suffer from long cycles and high costs,making it challenging to break through current limitations.Constructing computational models to predict potential miRNA-disease associations provides valuable assistance for biological experiments.In this paper,we propose network-based approaches to predict potential miRNA-disease associations using miRNA-disease similarity and association data.The specific contributions are as follows:(1)To address the problem of insufficient utilization of association and similarity information regarding miRNA and diseases,a method based on a two-layer weighted heterogeneous network is proposed to predict potential associations between miRNA and diseases.This method first defines the known associations between miRNA and diseases as different types of association pairs using a type definition strategy.Next,based on the proposed bidirectional information distribution strategy,the bidirectional affinity weights of each miRNA-disease association pair are calculated,allowing isolated nodes without any association information to participate in the prediction and fully utilize the similarity and association information of miRNA and diseases.Finally,the average of the two affinity weights is computed to obtain a comprehensive and accurate association score.(2)To overcome the issue of insufficient prior information for miRNA-disease association prediction,a method based on a three-layer heterogeneous network is proposed,which incorporates drug heuristic information and a bipartite network reconstruction strategy.Firstly,a three-layer heterogeneous network,consisting of miRNAs,drugs,and diseases,is constructed based on the association and similarity information among them.This increases the number of network paths from miRNA to disease nodes and enhances the complexity of the heterogeneous network.Secondly,a network path search algorithm is employed to calculate the direct and indirect path weights from miRNA to disease nodes,considering both internal and global perspectives within the network.This approach fully utilizes the similarity and association data of miRNAs,drugs,and diseases,enriching the prior information for prediction.Finally,the two types of paths are integrated to obtain a comprehensive miRNA-disease association prediction score.(3)To tackle the limitation of relying on a single similarity measure,we propose a miRNA-disease association prediction method based on a multi-view attention graph convolutional network.Firstly,we integrate multiple perspectives of miRNA-disease similarity information to construct multiple heterogeneous networks,ensuring comprehensive prior information.Additionally,we employ a graph convolutional neural network with layer attention mechanism for feature extraction of miRNA-disease associations,ensuring robust feature fusion and informative representation.Finally,we employ random forest as a predictor to further enhance the prediction performance of the model.