Prediction Of MiRNA And Disease Association Based On Graph Representation Learning

The development of biology and computer science promotes each other in the field of bioinformatics.The proportion of using the existing calculation methods to carry out biological related research is gradually increasing.Micro RNA(mi RNA)regulates the expression of various genes and ultimately affects the development of organisms.The occurrence of major human diseases is closely related to mi RNA mutation,Therefore,the successful prediction of pathogenic mi RNAs is important to the targeted treatment of various diseases and the exploration of the pathogenesis of diseases.The amount of biological network data is huge and complex,it is a key point to extract high-quality features by using complex mi RNA and disease association network to achieve more accurate prediction effect.Based on the above background,the main content of this paper is the prediction of mi RNA disease association.For the problem of sparsity of mi RNA and disease network,Due to the problem of calculation method,the original optimization method did not fit the original network characteristics.Therefore,this paper proposes a network optimization algorithm based on adding network similarity,which optimi zes the graph structure and makes the whole network more in line with the characteristics of natural network.The main idea of the algorithm is to calculate similarity network based on some tree medical topic words,and improve the integrity of the graph b y integrating disease network,mi RNA network,mi RNA family relationship and disease mi RNA relationship.The comparison experiments before and after the combination of the related networks prove that the method proposed in this paper is good for solving the sparse graph.In the process of network embedding,most of the fixed step size random walk can not extract the global network features better.In this paper,the original random walk method is improved to make the node sequence generated by random walk have more suitable global characteristics,and combine the global features with the local network features obtained by KL divergence,the final embedding effect is competitive with the mainstream methods such as LINE.In addition,for the problem that there is no high-quality negative sample in biological network,this paper also proposes a new negative case extraction method.Experiments show that the prediction effect of this method is good,and the accuracy of positive samples predicted for single disease mi RNA relationship is better,which can provide accurate research direction for researchers.