| Metabolic dysregulation is a hallmark of many diseases,including cancer,diabetes,cardiovascular disease,and Alzheimer’s disease.In view of the universal role of metabolism in various fields of pathology,it is of great significance to accurately describe metabolic changes and analyze the relationship between the differences in metabolic activities and diseases,which will contribute to research in the fields of clinical diagnosis,disease prevention and disease management.Although transcriptomics techniques have been used to analyze metabolic changes in diseases,most of the existing analysis methods tend to describe the average level of metabolic changes in heterogeneous cell subpopulations in tissues,and cannot accurately characterize changes of metabolic flux at the single-cell level.Single-cell transcriptome sequencing(scRNA-seq)data have been widely used to characterize the transcriptional state of cells in complex biological tissues.Transcriptional status is closely linked to metabolic changes,so scRNA-seq data has great application potential in metabolic research.Recently,some scholars have designed a computational model for estimating metabolic fluxes based on scRNA-seq data,which aims to analyze metabolic changes at the single-cell level.However,the metabolic flux estimated by the existing computational models cannot satisfy the flux balance constraints in the metabolic network,and the accuracy of the estimation needs to be further optimized.In addition,since the real metabolic flux is difficult to be directly measured,methods for testing and validating the metabolic flux estimation model need to be further explored.In order to estimate metabolic flux more precisely at the single-cell level,this thesis proposes a new model named sc FLUX for estimating metabolic flux based on scRNA-seq data.The model uses a Graph Neural Network(GNN)and a Message Passing Optimization algorithm(MPO)to estimate metabolic flux in the metabolic network.First,an initial value of metabolic flux is estimated using a self-supervised GNN.Next,the distribution of metabolic flux is optimized using MPO.Finally,a supervised fully-connected neural network is used to model the relationship between gene expression and metabolic flux,and the metabolic flux is further adjusted based on this relationship.This method models the metabolic network based on steady-state constraints,and by optimizing and adjusting the initial value of the metabolic flux distribution,it reduces the imbalance loss of metabolic flux in the network and improves the accuracy of estimation.In addition,this thesis proposes a method for modeling metabolic data based on Bounded Variable Least Squares to more comprehensively test and validate models for metabolic flux estimation.Finally,the thesis also presents online software developed based on these methods.In order to verify the effectiveness of the model,experiments were carried out on multiple metabolic networks closely related to human cancers compiled and reconstructed from the KEGG database.The gene expression data used in the experiments included two types: Transcriptome sequencing data related to dozens of cancers and other diseases collated from databases such as TCGA,CCLE and GEO,and simulated metabolic data generated by the metabolic data simulation method proposed in this thesis.The experimental results show that,compared with sc FEA and other comparison methods,sc FLUX has a significant improvement in the three indicators of IL,FC and EC.This shows that sc FLUX can effectively estimate the metabolic flux in the metabolic network using transcriptome sequencing data including scRNA-seq data,and the estimation accuracy of this model is superior to other methods.In addition,this thesis also verified the estimated results of sc FLUX through dozens of relevant literatures involving 12 cancers such as pancreatic cancer,gastric cancer,and liver cancer.The experimental data and conclusions given in many references support the estimated results of this model. |
PDF Full Text Request