Study On Component-target Interaction Relationship Based On Network Embeddin

Objective: Traditional Chinese medicine has a long clinical history and a remarkable curative effect,but little is known about the mechanisms of traditional Chinese medicine ingredients and therapeutic targets,posing significant challenges to clinical precision treatment and new drug research and development.This paper proposes an ingredient-target interaction analysis model based on heterogeneous network with the help of network embedding method to analyze the interaction between ingredients and targets of traditional Chinese medicine and alleviate the limitations of molecular structure and biochemical characteristics.Simultaneously,an improved graph convolutional neural network analysis model based on the characteristics of the heterogeneous network was proposed to investigate the interaction between traditional Chinese medicine ingredients and targets,providing a new way to analyze the modern material basis of traditional Chinese medicine.Methods: The information pertaining to the ingredients and targets of traditional Chinese medicine was sorted out,combined,and statistically examined before the data of traditional Chinese medicine were first acquired through the pharmacology database of traditional Chinese medicine system.Second,a heterogeneous network was built using the connections between the ingredients,targets,and traditional Chinese medicine.With the help of the XGBoost classifier and the ingredient-target interaction analysis model HIT2 vec based on heterogeneous network features,node features were extracted using the Node2 vec technique.Finally,the relationship perception module and meta-path aggregation module were built by multi-layer graph convolution to extract node features,and an improved graph convolution neural network model HITGCN is proposed to analyze the ingredient-target interaction relationship.These modules were constructed according to the characteristics of the heterogeneous network node types and relationship types.Results: According to the experimental results,the AUC and AUPR values of HIT2 vec are 90.9%and 93.3%,respectively,which are higher than those of the model using molecular attributes and the ingredient-target bipartite network.Simultaneously,the effect of the network embedding algorithm Node2 vec outperformed Line,Deepwalk,and Struc2 vec,and the effect of ensemble learning XGBoost as a classifier outperformed KNN,DNN,NB,and LR.The improved HITGCN model’s AUC and AUPR values for heterogeneous network characteristics were 93.6% and 94.5%,respectively,which were higher than the ablation experiment and the link prediction model of horizontal comparison.Furthermore,when compared to HIT2 vec,the AUC,F1,and AUPR of HITGCN increased by 2.7%,8.2%,and 1.2%,respectively.Conclusion: This paper proposes that the HIT2 vec model can effectively alleviate the influence of molecular structure and biochemical characteristics on feature extraction by constructing a heterogeneous network from the overall connection of traditional Chinese medicine,ingredients,and targets using network embedding technology.Based on the characteristics of the heterogeneous network,the improved graph convolutional neural network analysis model HITGCN with improved node representation ability is proposed.