Gene Markers Identification For Glioma Based On Single Cell Sequencing Data

Glioma accounts for about 50% of brain tumors and is with high mortality.However,the identification of glioma gene markers remains to be explored.The triggering factors of tumors include many types,in which the disturbance of gene expression level will directly affect bodily function.Single-cell gene expression data reflects gene expression level at a single-cell resolution,which helps to understand how potential molecular changes alter cell behaviors and disease processes.The disturbance of transcriptional regulatory relationships will also affect the gene expression level and then affect the occurrence and development of tumors.However,the influence of regulatory relationships has not been considered in many current studies of single-cell expression data.In this thesis,we fully took advantage of single-cell gene expression data and reasonably fused transcriptional regulatory relationships to propose an algorithmic framework for identifying tumor gene markers.The algorithm framework was applied to glioma,which could provide ideas for molecular mechanism and drug targeted therapy of glioma.The core elements of the algorithmic framework included: consensus genes identification,specific regulatory network construction,cell types identification with hybrid clustering and tumor marker genes identification.First,considering the obvious differences between different samples of tumor malignant cells,we first explored the expression status of malignant cells in a single sample through principal component analysis,cell-specific network construction,Louvain clustering,and differential genes identification.Then tumor consensus genes were identified by the overlap of differential genes among samples.The homology of consensus genes was analyzed from the perspective of genes and samples,the results showed that consensus genes reflected coexpression pattern of malignant cells in different samples.Then,in order to comprehensively analyze the characteristics of tumor,we combined multiple samples and rationally fused regulatory relationships to identify tumor cell types.Firstly,an initial regulatory network was constructed based on the regulatory relationships between transcription factors and target genes,then a specific regulatory network was constructed based on consensus genes,entire single-cell gene expression data and feed forward loop.Then based on the specific regulatory network,regulatory meta modules were identified and specific regulatory expression matrix was constructed.Finally,a hybrid clustering method was used in the specific regulatory expression matrix to identify the cell types of glioma,and the cell types were analyzed for enrichment analysis of gene ontology and biological pathway.The enrichment analysis indicated that the cell types had obvious functions,and the marker genes of cell types may be closely related to the tumor.Finally,the marker genes of cell types were as candidate genes,and a method for identifying tumor marker genes based on tumor eigen vector was proposed.In order to analyze the reliability of tumor marker genes,we analyzed two survival data of OS and PFI.Using tumor marker genes as classification features,seven binary classification algorithms were performed to construct risk prognostic classification models.The 14 types of risk prognostic classification models all showed good performance,of which the model with random forest performed best.Meanwhile,the tumor marker genes were analyzed by correlation measure,PubMed literature and Kaplan-Meier survival curve.The results showed that these genes were closely related to glioma.In addition,analysis of the results revealed that four genes(NDUFS5,NDUFA1,NDUFA13,and NDUFB8)in the tumor marker genes belong to the NADH ubiquinone oxidoreductase subunit gene family,indicating that this gene family may be closely related to glioma.The experimental results showed the effectiveness of the proposed algorithm framework,revealed six cell types of glioma malignant cells,and predicted 20 tumor marker genes,which were of great significance for pathological mechanism and precise treatment of glioma.