Research On Representation Learning Of Human Long Non-coding RNA-disease Relationship Network

As the research on the relationship between human long non-coding RNA and disease continues,methods for predicting long non-coding RNA-disease relationship have emerged.The proven long non-coding RNA and disease relationship is introduced to form a relationship network.Researchers use network representation learning to obtain node representation vectors and use it to predict long non-coding RNA-disease relationships.The node representation vector can play a good role in network correlation analysis.In this project,we use long non-coding RNA and disease relationship data to build a two-layer heterogeneous network composed of 190 long non-coding RNA and disease nodes.This network contains 3633 undirected edges.However,based on the network relationship data of this project,the homogeneous network representation learning method does not add community constraint information and the heterogeneous network representation learning method does not introduce similarity node similarity data.The representation learning method of homogeneous networks has no community constraints.After performing homogeneous input processing on long non-coding RNA-disease relationship heterogeneous networks,we performed community discovery on long non-coding RNA similarity networks and disease similarity networks,improved community constraints,and experimental analysis of network relationship prediction.Heterogeneous networks indicate that there is a problem in learning methods that cannot take advantage of long non-coding RNA and disease similarity data.On the long non-coding RNA-disease relationship heterogeneous network,we introduce community members to increase jump objects,combine the improvement of similarity networks,and experimental analysis of network relationship prediction.The relationship prediction results obtained by the two improved representation learning methods are compared with the restart random walk RWR method,the Deepwalk representation learning method,and the Node2 vec representation learning method.In addition,the improved heterogeneous network representation learning method combined with the similarity network is compared with the Metapath2 vec representation learning method.In this paper,the relationship prediction results of the homogeneous network representation learning method with improved community constraints are improved by 1.65% over the Deepwalk representation learning method and 3.12% over the Node2 vec representation learning method.The relationship prediction results of the heterogeneous network representation learning method combined with the similarity network improvement are 3.34% higher than the Deepwalk representation learning method,4.81% higher than the Node2 vec representation learning method,and 6.29% higher than the Metapath2 vec representation learning method.Finally,this paper uses nodes to represent vectors to calculate the cosine similarity.The calculation results are verified by the literature,indicating that the method of this topic can guide the experiment.