Integrative bulk and single-cell transcriptomic profiling identifies core gene networks and potential therapeutic targets in glioma Authors: Xingxing Jiang, Yongping Gui, Yu Yang, Jie Li, Aijun Liang Published: 2025 ABSTRACT AbstractBackgroundGlioma is the most common primary malignant tumor of the central nervous system and is associated with an extremely poor prognosis. Given its highly complex molecular landscape, there is an urgent need to identify novel diagnostic biomarkers. This study systematically identified core glioma genes through multi-cohort integration and single-cell analysis.MethodsFive Gene Expression Omnibus (GEO) transcriptomic datasets (GSE109857, GSE15824, GSE35158, GSE4290, and GSE90886; 504 samples) were integrated, and batch effects were corrected using ComBat. Differentially expressed genes (DEGs) were identified and combined with weighted gene co-expression network analysis (WGCNA) to obtain candidate modules. Feature genes were further screened using multiple approaches, including CytoHubba, least absolute shrinkage and selection operator (LASSO) regression, support vector machine-recursive feature elimination (SVM-RFE), and random forest (RF) analysis. Validation was performed using external GEO datasets, receiver operating characteristic (ROC) analysis, and quantitative real-time polymerase chain reaction (qRT-PCR) based on 69 clinical samples. Additional analyses included gene set enrichment analysis (GSEA), CIBERSORT-based immune infiltration profiling, expression mapping across three single-cell RNA sequencing (scRNA-seq) datasets (GSE273274, GSE162631, and GSE182109), and construction of competing endogenous RNA (ceRNA) and transcription factor (TF) networks, as well as drug enrichment analysis.ResultsA total of 409 glioma-related candidate genes were identified. Six hub genes, SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1, were consistently identified across all algorithms. All six genes were significantly downregulated in glioma, as confirmed by GEO datasets (area under the curve [AUC]: 0.787–0.802) and clinical qRT-PCR validation. GSEA revealed that low expression of these genes was associated with activation of the cell cycle, complement and coagulation cascades, and immune dysregulation. Immune infiltration analysis showed negative correlations with M1/M2 macrophages and activated natural killer (NK) cells. Single-cell analyses indicated that these hub genes were primarily enriched in tumor-propagating cell (TPC)-like tumor cells but were markedly reduced in glioma core regions. Competing endogenous RNA (ceRNA)/TF networks and drug enrichment analyses suggested multilayered regulatory mechanisms and associations with neurotransmitter-related compounds.ConclusionThe six identified hub genes are consistently downregulated in glioma and exhibit strong diagnostic potential. Their close association with the immune microenvironment and tumor-cell lineage highlights their value as biomarkers and potential therapeutic targets.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-025-15454-5.Keywords: Glioma, Core genes, WGCNA, Machine learning, Immune infiltration, Single-cell transcriptome, CeRNA network, Transcription factors, Drug enrichment analysis Background Glioma is one of the most common primary malignant tumors of the central nervous system, accounting for approximately 30% of all brain tumors. It is characterized by high malignancy, pronounced heterogeneity, and an extremely poor prognosis [1, 2]. Among them, glioblastoma multiforme (GBM) is the most aggressive subtype, exhibiting rapid growth and marked invasiveness, with a median survival of less than 15 months and a five-year survival rate of under 5% [3, 4]. Although surgical resection combined with radiochemotherapy and molecular targeted therapies (e.g., temozolomide and bevacizumab) has become the current clinical standard, the overall therapeutic efficacy remains unsatisfactory. These observations suggest that the underlying molecular mechanisms of glioma pathogenesis and progression are still not fully elucidated [5, 6]. In recent years, extensive transcriptomic and genomic studies have revealed key molecular events and dysregulated signaling pathways in glioma. Molecular biomarkers such as isocitrate dehydrogenase (IDH) mutations, 1p/19q co-deletion, telomerase reverse transcriptase (TERT) promoter mutations, and O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation have been confirmed to be closely associated with tumorigenesis, disease progression, and patient prognosis [7–12]. However, due to the pronounced heterogeneity of gliomas, tumor development is not solely driven by mutations in oncogenes or tumor suppressor genes, but also by the interplay of complex gene regulatory networks, immune microenvironment remodeling, and neural signaling disruption [13–15]. Therefore, elucidating the core molecular networks from a systems biology perspective and identifying novel biomarkers and potential therapeutic targets are of great importance. In this study, we integrated multiple publicly available transcriptomic datasets and employed differential expression analysis in combination with weighted gene co-expression network analysis (WGCNA) to identify candidate genes associated with glioma. Core network genes were further refined using machine learning algorithms, and their expression patterns were validated in clinical glioma tissue samples, demonstrating strong concordance with the bioinformatic predictions. To investigate the biological functions of these core genes and their relevance to the tumor immune microenvironment, we performed gene set enrichment analysis (GSEA), immune infiltration profiling, and single-cell transcriptomic analysis. In addition, competing endogenous RNA (ceRNA) network construction, transcription factor (TF) prediction, and drug enrichment analysis were conducted to explore upstream regulatory mechanisms and potential therapeutic implications. Collectively, this study provides new insights into the molecular mechanisms underlying glioma and establishes a theoretical framework for the identification of novel diagnostic biomarkers and therapeutic strategies. Materials and methods Data acquisition and processing Five publicly available glioma transcriptome datasets were retrieved from the Gene Expression Omnibus (GEO) database: GSE109857 (5 non-tumor and 220 tumor samples), GSE15824 (2 non-tumor and 30 tumor samples), GSE35158 (1 non-tumor and 80 tumor samples), GSE4290 (23 non-tumor and 157 tumor samples), and GSE90886 (9 non-tumor and 9 tumor samples). After downloading the raw data, probe annotation files were curated and uniformly mapped to official gene symbols. For genes represented by multiple probes, the average expression value was calculated as the final gene expression level. To enhance comparability across datasets, batch effects were corrected using the “ComBat” method implemented in the “sva” R package, ensuring the reliability of subsequent analyses. Differentially expressed genes (DEGs) were identified using the “limma” R package, with the thresholds set at |log2FC| >1 and false discovery rate (FDR) < 0.05 for the comparison between glioma and normal tissues. WGCNA To systematically identify gene modules associated with glioma, WGCNA was performed to construct weighted co-expression networks of DEGs [16–20]. The main steps included: (i) calculating Pearson correlation coefficients between genes to generate a similarity matrix; (ii) selecting an appropriate soft-thresholding power (β value) according to scale-free topology criteria to transform the similarity matrix into an adjacency matrix; (iii) computing the topological overlap matrix (TOM) and performing hierarchical clustering to define gene modules; and (iv) correlating module eigengenes with clinical traits (tumor vs. normal status) to identify the modules most strongly associated with glioma. Candidate gene selection and functional enrichment analysis The intersection of DEGs and genes derived from key WGCNA modules was defined as the candidate gene set. To investigate their potential biological significance, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis and Disease Ontology (DO) analysis were performed using the “clusterProfiler” R package, thereby identifying enriched pathways and disease associations. In addition, a protein-protein interaction (PPI) network was constructed using the STRING database and subsequently visualized in Cytoscape. Multi-method identification of network core genes Within the PPI network, node genes were ranked using the CytoHubba plugin in Cytoscape, and potential hub genes were identified using four topological algorithms: EPC, MCC, MNC, and Degree [21]. To further refine the selection and improve robustness, three machine learning approaches were applied for feature selection and validation: Least absolute shrinkage and selection operator (LASSO) regression: Conducted using the “glmnet” R package, with cross-validation to determine the optimal λ value and obtain a stable set of key genes. Support Vector Machine Recursive Feature Elimination (SVM-RFE): Iteratively removed non-informative features, while calculating model accuracy and cross-validation errors across different feature subsets to identify the optimal gene set. Random Forest (RF) analysis: Performed using the “randomForest” R package, with gene importance scores computed to rank candidate genes. The intersection of results from these three methods was considered as the final high-confidence set of network core genes. Expression levels of these genes were first validated across GEO glioma cohorts, and their diagnostic performance in distinguishing tumor from normal tissues was further assessed by calculating the area under the receiver operating characteristic (ROC) curve (AUC). Diagnostic performance evaluation and nomogram construction ROC curves and the corresponding AUC values were calculated with the ‘pROC’ R package to assess the diagnostic performance of the six-gene model. Calibration curves were generated to evaluate the agreement between predicted and observed probabilities of glioma. Decision curve analysis (DCA) was further used to quantify the net clinical benefit of the model across a spectrum of threshold probabilities. In addition, a nomogram incorporating the six hub genes was constructed using the ‘rms’ R package to facilitate individualized risk prediction in clinical practice. Clinical sample collection and quantitative real-time PCR (qRT-PCR) validation To further validate the results of the bioinformatics analysis, clinical glioma samples were collected from Jiangxi Provincial People’s Hospital, including 69 pairs of glioma tissues and adjacent non-tumor tissues. All samples were obtained with informed consent from patients and approved by the Ethics Committee of Jiangxi Provincial People’s Hospital (Approval No.:2020-066). Total RNA was extracted from tissues using TRIzol reagent (Invitrogen, USA), and RNA concentration and purity were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA). Complementary DNA (cDNA) was synthesized using the PrimeScript RT Reagent Kit (Takara, Japan), followed by quantitative PCR amplification with SYBR Green PCR Master Mix (Takara, Japan), according to the manufacturer’s instructions. GAPDH served as the internal control, and relative gene expression levels were calculated using the 2−ΔΔCt method. Primers for network core genes and GAPDH were synthesized by Sangon Biotech (Shanghai, China). The primer sequences were as follows: GAPDH, Forward: 5’-GGAGCGAGATCCCTCCAAAAT-3’, Reverse: 5’-GGCTGTTGTCATACTTCTCATGG-3’; SYP, Forward: 5’-CTCGGCTTTGTGAAGGTGCT-3’, Reverse: 5’-CTGAGGTCACTCTCGGTCTTG-3’; SYN1, Forward: 5’-AGTTCTTCGGAATGGGGTGAA-3’, Reverse: 5’-CAAACTGCGGTAGTCTCCGTT-3’; RAB3A, Forward: 5’-GAGTCCTCGGATCAGAACTTCG-3’, Reverse: 5’-TGTCGTTGCGATAGATGGTCT-3’; SLC17A7, Forward: 5’-CTGGGGCTACATTGTCACTCA-3’, Reverse: 5’-GCAAAGCCGAAAACTCTGTTG-3’; SYN2, Forward: 5’-CCCAACGGCTACATGACCG-3’, Reverse: 5’-CCGTCTGCTTTACCGCTTG-3’; STXBP1, 5’-AAAGCTGTTGTCGGAGAGAAG-3’, Reverse: 5’-CACAATCGTTATGCCCTCGG-3’. GSEA For each core gene, samples in each cohort were dichotomized into high- and low-expression groups according to the cohort-specific median expression level of that gene. Samples with expression values greater than or equal to the median were assigned to the high-expression group, whereas those with expression values below the median were assigned to the low-expression group. GSEA was then performed by comparing the high- versus low-expression groups for each core gene. Immune cell infiltration analysis The CIBERSORT algorithm was applied to estimate the relative proportions of 22 immune cell types in both normal brain tissues and glioma tissues. Differences in immune cell composition between the two groups were evaluated. In addition, correlations between the expression levels of the core genes and the degree of immune cell infiltration were calculated to assess the potential roles of these genes within the tumor immune microenvironment. Preprocessing of single-cell RNA sequencing (scRNA-seq) data Publicly available scRNA-seq datasets of glioma were obtained from the GEO database, including GSE273274 (3 tumor-core and 3 tumor-periphery samples), GSE162631 (4 tumor-core and 4 tumor-periphery samples), and GSE182109 (4 tumor-core samples). The raw single-cell FASTQ files were aligned to the GRCh38 reference genome using Cell Ranger (version 7.2, 10x Genomics) with default parameters. Count matrices were processed and analyzed using the Python package Scanpy (version 1.9.8). Quality control was performed to remove low-quality cells and potential doublets. Cells with fewer than 200 or more than 6,000 detected features, as well as those with mitochondrial gene expression exceeding 25% of total unique molecular identifier (UMI) counts, were excluded. Only genes detected in at least three cells were retained. Doublet detection was conducted using the Scrublet algorithm, which simulates artificial doublets to estimate a doublet score for each cell; cells identified as high-confidence doublets based on Scrublet thresholds were removed. Normalization was performed using scanpy.pp.normalize_total (target_sum = 1e4) followed by scanpy.pp.log1p. The count matrix was normalized to 10,000 counts per cell according to total UMI counts, log-transformed by adding one, and then standardized for each gene. Then, variable genes were identified based on defined thresholds for mean expression and dispersion using the scanpy.pp.highly_variable_genes function. Principal component analysis (PCA) was performed on the normalized expression matrix of highly variable genes, retaining the top 50 principal components (PCs). Batch correction was applied using Scanpy’s Harmony and Ingest functions with the PC model, generating batch-corrected PCs for downstream analyses. Uniform Manifold Approximation and Projection (UMAP) was subsequently used to visualize latent cell-state embeddings. Batch correction and data integration To eliminate batch effects arising from different donors, sequencing runs, or experimental conditions, data integration was performed using single-cell variational inference (scVI, scvi-tools). The raw count matrix, together with batch labels, was used to train the scVI model. Cell-type annotation Marker genes for each Leiden cluster were identified using the scanpy.tl.rank_genes_groups function, which performs differential expression analysis between a given cluster and all remaining cells. DEGs were calculated using the Wilcoxon rank-sum test. Cluster-specific DE genes were then compared against canonical marker genes curated from previous studies and publicly available cell-type reference databases. A cluster was assigned a cell identity if it showed consistent enrichment of well-established markers for that lineage. Cell-type annotation was additionally guided by the original cell labels provided in the GSE273274 dataset [22]. Inference of TF activity TF activity was inferred from the integrated single-cell RNA sequencing dataset using the decoupler-python framework (version 1.9.2), which implements multiple gene regulatory network-based enrichment approaches. TF regulon activity scores were computed for each cell based on curated TF-target gene interactions. TFs with low or non-significant inferred activity, or lacking meaningful associations with the expression of their predicted target genes, were excluded. The remaining TFs with robust inferred activity were used to construct the upstream TF-core gene regulatory network. ceRNA network Based on the ceRNA hypothesis, interactions between core genes and microRNAs (miRNAs) were predicted by integrating TargetScan, miRanda, and miRDB databases. To obtain a high-confidence set of interactions, only miRNA–mRNA pairs consistently predicted by all three databases were retained. Upstream long non-coding RNAs (lncRNAs) were further predicted using the spongeScan database, and an lncRNA–miRNA–mRNA regulatory network was constructed. Drug enrichment analysis To explore the potential clinical applications of the core genes, drug enrichment analysis was conducted using the Drug Signatures Database (DSigDB). Small-molecule compounds and chemicals significantly associated with the core genes were identified, annotated with their biological functions, and evaluated for their potential as therapeutic targets or predictive markers of drug sensitivity. Statistical analysis All statistical analyses were performed using R software (v4.3.3) and GraphPad Prism 8.0. Data were expressed as mean ± standard deviation (SD). Differences between the two groups were assessed using Student’s t-test. Correlations were analyzed using Pearson correlation coefficients. ROC curves and the corresponding AUC values were calculated with the “pROC” R package. All statistical tests were two-sided, and P < 0.05 was considered statistically significant. Results Identification of glioma core genes by integrating WGCNA and differential expression analysis In this study, we integrated five publicly available transcriptomic datasets (GSE109857, GSE15824, GSE35158, GSE4290, and GSE90886), comprising 504 samples (non-tumor = 40; tumor = 464). Because different platforms and experimental batches may introduce systematic biases, we first applied ComBat to correct batch effects across the merged gene expression matrix. PCA revealed clear inter-cohort separation before correction, whereas samples showed substantial convergence after correction (Fig. 1a), indicating that batch effects were largely eliminated. Integrated identification of glioma-related DEGs and WGCNA modules, and pathway enrichment analysis. (a) PCA plots before and after batch effect correction. (b) Volcano plot of differential expression analysis. (c) Soft-threshold power selection in WGCNA. (d) Heatmap of correlations between WGCNA modules and clinical traits. (e) Venn diagram of DEGs and WGCNA module genes. (f) KEGG pathway enrichment bubble plot. (g) Disease Ontology (DO) enrichment bar plot Differential expression analysis identified numerous genes significantly altered in glioma (Fig. 1b, Table S1). To further determine disease-relevant co-expression patterns, we performed WGCNA. Before constructing the network, we evaluated key quality-control metrics, including the scale-free topology fit index (R2) and mean connectivity across multiple soft-thresholding powers (Fig. 1c). As the soft-threshold increased, R2 rose steadily and approached stabilization near 0.9, while mean connectivity declined. These results indicated that a soft-thresholding power of 7 achieved scale-free topology while minimizing noise. Based on gene expression similarity, a weighted co-expression network was constructed, and initial modules were identified using dynamic tree cutting. Modules with highly correlated eigengenes were subsequently merged (Figure S1a-d). Among all resulting modules, the turquoise module exhibited the strongest correlation with the glioma phenotype (Fig. 1d) and displayed the highest gene significance (GS) values (Figure S1e), suggesting that this module plays a crucial biological role in glioma development. To identify core functional genes genuinely associated with glioma, we intersected the DEGs with those in the turquoise module, yielding 409 candidate key genes (Fig. 1e; Table S2). Functional enrichment analysis demonstrated that these genes were significantly enriched in multiple neural-related pathways, including neurotransmitter receptor interactions, calcium signaling, GABAergic and glutamatergic synapses, and long-term potentiation (Fig. 1f). DO enrichment further revealed strong associations with various neurological and psychiatric disorders, such as autism spectrum disorder, epilepsy, schizophrenia, and depression (Fig. 1g), highlighting their involvement in neural regulatory processes. A PPI network constructed using the STRING database (Figure S2) demonstrated dense interaction patterns among the candidate genes, indicating that they may constitute key functional hubs regulating glioma-related signaling pathways. Collectively, by integrating differential expression analysis and WGCNA, we systematically identified 409 glioma-associated candidate hub genes, providing a solid foundation for subsequent mechanistic investigations and the exploration of potential therapeutic targets. Screening of network core genes in glioma To further refine the candidate genes and identify network core genes involved in glioma progression, nodes in the PPI network were first scored using four topological algorithms (EPC, MCC, MNC, and Degree) implemented in CytoHubba, resulting in a set of potential hub genes (Fig. 2a, Table S3). Subsequently, LASSO regression was performed for feature selection. Cross-validation demonstrated that an optimal λ value yielded a stable set of key genes (Fig. 2b-c). In parallel, SVM-RFE analysis indicated that the model achieved the highest accuracy and lowest cross-validation error when the number of features was set to 10 (Fig. 2d-e). Furthermore, RF analysis ranked candidate gene importance, with SLC17A7, SYN1, and SYP among the top-ranked genes (Fig. 2f). By integrating the results from LASSO, SVM-RFE, and RF analyses, six overlapping genes were identified: SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1 (Fig. 2g). Screening of network core genes in glioma. Screening of network core genes in glioma. (a) Venn diagram showing candidate hub genes identified by four CytoHubba algorithms (EPC, MCC, MNC, and Degree). (b-c) Results of the least absolute shrinkage and selection operator (LASSO) regression analysis. (b) Selection of the optimal λ parameter under cross-validation; (c) regression coefficient profiles of candidate genes across different λ values. (d-e) Results of support vector machine–recursive feature elimination (SVM-RFE). (d) Ten-fold cross-validation accuracy for different feature numbers; (e) cross-validation error curve, with the optimal feature number determined as 10. (f) Gene importance ranking from random forest (RF) analysis, with SLC17A7, SYN1, and SYP among the top-ranked genes. (g) Venn diagram showing six overlapping genes identified by LASSO, SVM, and RF analyses. (h) Boxplots showing expression levels of candidate genes between normal and glioma tissues. SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1 were significantly downregulated in glioma (***P < 0.001). (i) Receiver operating characteristic (ROC) curve analysis demonstrating the diagnostic performance of the six genes in distinguishing glioma from normal tissues, with area under the curve (AUC) values ranging from 0.787 to 0.802. (j) Quantitative real-time PCR (qRT-PCR) validation of mRNA expression levels of the six hub genes in 69 pairs of glioma and adjacent non-tumor tissues. (k) Calibration curve of the diagnostic model, evaluating the agreement between predicted and observed probabilities. (l) Decision curve analysis (DCA) showing the net clinical benefit of the model across a range of threshold probabilities. (m) Nomogram constructed from the six network hub genes, enabling individualized prediction of glioma risk. Expression analysis revealed that all six genes were significantly downregulated in glioma tissues compared with normal tissues (Fig. 2h). ROC curve analysis further confirmed their diagnostic potential, with AUC values ranging from 0.787 to 0.802 (Fig. 2i), suggesting good diagnostic performance. To validate these findings experimentally, qRT-PCR was performed on 69 paired glioma and adjacent normal tissues. Consistent with the bioinformatic predictions, all six core genes were markedly downregulated in tumor tissues (Fig. 2j). Building on these findings, we constructed a predictive model based on the six network hub genes. Calibration curves were generated to assess the agreement between predicted and observed probabilities, demonstrating excellent concordance (Fig. 2k). In addition, decision curve analysis (DCA) showed that the model provided substantial net clinical benefit across a broad range of threshold probabilities (Fig. 2l). Finally, a nomogram incorporating the six hub genes was established to facilitate individualized risk assessment, enabling quantitative prediction of glioma occurrence (Fig. 2m). Taken together, SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1 were identified as persistently downregulated network hub genes in glioma. These genes exhibit not only robust biological relevance but also significant potential for clinical diagnosis and risk prediction. GSEA enrichment analysis of network core genes To further elucidate the potential biological functions of the six diagnostic-related core genes (SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1) in glioma, GSEA was performed. Representative significantly enriched pathways for each core gene are shown in Fig. 3a-f, and the complete GSEA results are summarized in Supplementary Table S4-S9. The results showed that in samples with low RAB3A expression, significantly enriched pathways included cell cycle, complement and coagulation cascades, cytokine-cytokine receptor interaction, and leishmaniasis (Fig. 3a). In samples with low SLC17A7 expression, enriched pathways were associated with cell cycle, DNA replication, complement and coagulation cascades, and extracellular matrix (ECM)–receptor interaction (Fig. 3b). For STXBP1, low-expression samples were predominantly enriched in immune-related pathways, such as graft-versus-host disease, autoimmune thyroid disease, and complement and coagulation cascades (Fig. 3c). Similarly, low SYN1 expression was associated with enrichment in graft-versus-host disease, autoimmune thyroid disease, and complement and coagulation cascades (Fig. 3d). In low SYN2 expression samples, enrichment was primarily observed in cell cycle, complement and coagulation cascades, and cytokine-cytokine receptor interaction (Fig. 3e). In low SYP expression samples, significantly enriched pathways included autoimmune thyroid disease, complement and coagulation cascades, and cytokine-cytokine receptor interaction (Fig. 3f). Conversely, samples with high expression of the core genes were enriched primarily in neural and metabolic pathways, including calcium signaling, cardiac muscle contraction, long-term potentiation, neuroactive ligand-receptor interaction, and oxidative phosphorylation (Figure S3). Gene set enrichment analysis (GSEA) enrichment analysis of key candidate genes in glioma. (a) In samples with low RAB3A expression, significantly enriched pathways included cell cycle, complement and coagulation cascades, and cytokine–cytokine receptor interaction. (b) In samples with low SLC17A7 expression, enriched pathways included cell cycle, complement and coagulation cascades, and DNA replication. (c) In samples with low STXBP1 expression, enriched pathways included graft-versus-host disease, autoimmune thyroid disease, and complement and coagulation cascades. (d) In samples with low SYN1 expression, enriched pathways included graft-versus-host disease, autoimmune thyroid disease, and complement and coagulation cascades. (e) In samples with low SYN2 expression, enriched pathways included cell cycle, complement and coagulation cascades, and cytokine–cytokine receptor interaction. (f) In samples with low SYP expression, enriched pathways included autoimmune thyroid disease, complement and coagulation cascades, and cytokine-cytokine receptor interaction. Taken together, these results indicate that low expression of the six core candidate genes is consistently associated with cell cycle dysregulation, abnormal immune responses, and activation of the complement and coagulation systems. Correlation analysis between network core genes and immune cell infiltration To further explore the roles of the network core genes in the tumor immune microenvironment, the CIBERSORT algorithm was used to estimate the relative proportions of 22 immune cell types in normal brain tissues and glioma tissues (Fig. 4a). The results revealed marked differences in immune cell distribution between the two groups, with M1 macrophages significantly enriched in glioma tissues, while activated dendritic cells were markedly suppressed (Fig. 4b). Correlation analysis between network core genes and immune cell infiltration in glioma. (a) Relative proportions of 22 immune cell types in normal and glioma tissues inferred by the CIBERSORT algorithm. (b) Boxplots showing differences in immune cell fractions between normal and glioma tissues, indicating significant enrichment of certain immune cells in tumors. (c-h) Correlations between network core genes and immune cell infiltration levels. (c) RAB3A was significantly negatively correlated with multiple immune cell types, particularly M1/M2 macrophages and activated NK cells. (d) SLC17A7 was negatively correlated with M1 macrophages, activated NK cells, and follicular helper T cells. (e) STXBP1 showed significant negative correlations with M1/M2 macrophages and activated NK cells. (f) SYN1 was negatively correlated with activated NK cells, M1 macrophages, and follicular helper T cells. (g) SYN2 showed significant negative correlations with M1/M2 macrophages and NK cells. (h) SYP was significantly negatively correlated with M1/M0 macrophages and activated NK cells. Dot size represents the absolute value of correlation coefficients, while color indicates statistical significance (p-value) Correlation analysis further demonstrated that all six core genes were significantly associated with immune cell infiltration. Specifically, RAB3A was negatively correlated with multiple immune cell types, particularly M1/M2 macrophages and activated natural killer (NK) cells (Fig. 4c). SLC17A7 expression was negatively correlated with M1 macrophages, activated NK cells, and follicular helper T cells (Fig. 4d). STXBP1 also showed negative correlations with M1/M2 macrophages and activated NK cells (Fig. 4e). Similarly, SYN1 expression exhibited significant negative correlations with activated NK cells, M1 macrophages, and follicular helper T cells (Fig. 4f). SYN2 was negatively correlated with M1/M2 macrophages and activated NK cells (Fig. 4g). SYP also demonstrated significant negative correlations with M0/M1 macrophages and activated NK cells (Fig. 4h). Taken together, these findings suggest that low expression of the six core candidate genes is not only linked to cell cycle dysregulation and immune-related pathway alterations (Fig. 3), but may also contribute to glioma initiation and progression by modulating the tumor immune microenvironment, particularly through effects on macrophage polarization and NK cell activation. Single-cell transcriptome analysis reveals the distribution of network core genes across cell subpopulations To further characterize the expression patterns of the network hub genes across distinct cellular populations in glioma, we integrated three scRNA-seq datasets (GSE273274, GSE162631, and GSE182109) to obtain a unified, high-quality single-cell transcriptomic dataset. As shown in Figure S4a, key quality-control metrics, including total UMI counts, number of detected genes (n_genes_by_counts), percentage of mitochondrial transcripts (pct_counts_mt), and doublet_score, displayed reasonable distributions across cells. Cells with excessively high mitochondrial transcript proportions or predicted doublets were removed, ensuring data suitability for downstream analyses. Following dimensionality reduction and batch correction using PCA, followed by scVI/Harmony integration, both t-SNE and UMAP consistently revealed stable and reproducible clustering patterns (Figure S4b). A total of 11 major cell populations were identified, including AC-like, MES-like, NPC-like, OPC-like, TPC-like, Oligo, EC, Macro, MC, Lym, and Neutrophils. These cell types formed distinct low-dimensional clusters (Fig. 5a), and their representative marker genes exhibited characteristic expression profiles, further validated by DotPlot analysis (Fig. 5b). The strong concordance between UMAP- and t-SNE-based annotations underscores the robustness of the cell-type classification. Single-cell transcriptomic analysis reveals the distribution of network hub genes across glioma cell subpopulations. Single-cell transcriptomic analysis reveals the distribution of network hub genes across glioma cell subpopulations. (a) Uniform Manifold Approximation and Projection (UMAP) visualization of the integrated single-cell RNA sequencing (scRNA-seq) data from three datasets (GSE273274, GSE162631, and GSE182109), showing 11 major cell populations, including AC-like, EC, Lym, MC, MES-like, Macro, NPC-like, Neutrophils, OPC-like, Oligo, and TPC-like cells. Representative marker genes for each cluster are indicated in the outer ring. (b) DotPlot of canonical marker genes across the identified cell subpopulations, demonstrating characteristic expression signatures and validating the accuracy of the cell-type annotations. (c) DotPlot showing the expression of the six network hub genes across different cell populations. Color intensity represents the average expression level, and dot size indicates the percentage of cells expressing the gene. (d) Violin plots illustrating expression differences of the network hub genes between the tumor core (Tumor) and tumor periphery In the UMAP embedding, the two tissue origins displayed pronounced differences in cellular composition (Figure S4c). Tumor core samples were enriched for classical malignant populations, including MES-like, NPC-like, TPC-like, and OPC-like tumor cells. In contrast, tumor periphery samples contained a higher proportion of immune populations (Macro, MC, and Lym), along with subsets of EC and Oligo cells, forming a spatially stratified cellular architecture. Based on this integrated cellular atlas, we next examined the expression of the six network hub genes across the identified cell populations (Fig. 5c). The results showed that the hub genes were predominantly expressed in TPC-like tumor cells, whereas their expression was relatively low across other cell types. Further comparison of gene expression between tumor core and tumor periphery (Fig. 5d) revealed that many of these genes were markedly downregulated within malignant subpopulations in the tumor core but showed elevated expression in corresponding cell populations at the periphery. These results indicate that the hub genes exhibit distinct spatial expression patterns, reflecting functional differences between malignant cell lineages and the immune-rich microenvironment at the tumor margins. CeRNA regulatory Network, TF Regulation, and drug enrichment analysis To further explore the upstream regulatory mechanisms of the network core genes, a potential ceRNA network was constructed. The results showed that core genes, including SYP, SYN1, RAB3A, SLC17A7, and STXBP1 (red nodes), interacted with multiple miRNAs (green nodes), which in turn competitively bound upstream lncRNAs (blue nodes), forming a complex lncRNA-miRNA-mRNA regulatory network (Figure S5). These findings suggest that the network core genes may contribute to glioma progression through post-transcriptional regulation mediated by the ceRNA mechanism. Furthermore, TF activity was inferred from the integrated single-cell dataset using the decoupler-python framework, and an upstream regulatory network of feature genes was constructed specifically within the TPC-like subpopulation (Fig. 6a). The results revealed that several key TFs exhibited markedly elevated activity in TPC-like cells and formed tightly interconnected regulatory relationships with the network’s core feature genes. Subsequent correlation analysis (Fig. 6b) demonstrated that the inferred TF activity levels were significantly positively or negatively correlated with the expression of their corresponding target genes, indicating clear regulatory directionality within this cellular subpopulation. Transcription factor (TF) activity inference using decoupler-python reveals upstream regulatory networks of the feature genes.Transcription factor (TF) activity inference using decoupler-python reveals upstream regulatory networks of the feature genes. (a) TF regulatory network inferred by the decoupler-python framework within the TPC-like cell population. Square nodes represent inferred upstream TFs (sources), and circular nodes represent feature genes (targets). (b) Scatter plots showing the correlations between TF activity and target gene expression within the TPC-like subpopulation, illustrating regulatory associations between TF activity levels and downstream gene expression To assess the potential clinical significance of these genes, drug enrichment analysis was performed using DSigDB. The results revealed significant associations between the candidate genes and a range of small molecules, including neurotransmitter-related compounds such as γ-aminobutyric acid (GABA), N-methyl-D-aspartic acid (NMDA), fluoxetine hydrochloride, and L-glutamic acid, as well as environmental chemicals such as chlorpyrifos oxon and aluminium sulfate (Figure S6). These findings suggest that the core genes may be involved in neurotransmitter signaling and responses to exogenous chemicals, highlighting their potential as drug targets or biomarkers of drug sensitivity. Collectively, these findings indicate that the six core candidate genes may contribute to glioma pathogenesis not only through ceRNA- and TF-mediated regulation but also by interacting with neurotransmitter-related compounds and small molecules, thereby providing novel insights for functional studies and targeted therapeutic strategies. Discussion By integrating five transcriptomic datasets and combining differential expression analysis with WGCNA, this study systematically identified six key genes (SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1) that were consistently downregulated in glioma. These genes are broadly involved in synaptic transmission, neuronal development, and neural network regulation. Their pronounced suppression suggests that glioma cells may undergo profound “neuronal phenotype loss” accompanied by extensive synaptic functional reprogramming. Several methodological considerations should also be acknowledged. First, clinical metadata such as age, sex, and treatment information were not uniformly available or reliable across the five public cohorts. Therefore, tumor versus non-tumor status was used as the primary trait in WGCNA, which may limit the ability to disentangle the potential influences of these covariates. Future studies incorporating datasets with comprehensive and harmonized clinical annotations will be essential to refine the interpretation of the identified co-expression modules. It is noteworthy that although the correlation between the turquoise module and the glioma phenotype in this study did not reach traditionally high thresholds, similar situations are frequently observed in large-cohort, multi-platform integrative analyses. For example, both the Dupilumab resistance study (r = 0.47) and the Bronchopulmonary Dysplasia study (r = 0.52) regarded modules with comparable correlation strengths as key modules [23, 24]. This phenomenon largely reflects cohort heterogeneity, cross-platform technical noise, and the intrinsic nature of WGCNA, which captures patterns of gene co-expression rather than linear effect sizes. Despite the moderate correlation, the turquoise module exhibited the highest gene significance, showed substantial overlap with the GBMs, and was enriched in canonical neural and glioma-related pathways. Taken together, these multiple lines of evidence strongly support the biological relevance of this module, establishing it as a critical foundation for identifying the core genes in this study. Synaptophysin (SYP), a synaptic vesicle protein, exhibited a “higher grade, lower expression” pattern in glioma, and its downregulation was significantly associated with poor survival outcomes, suggesting potential utility as a molecular biomarker for tumor grading and prognosis [25, 26]. Synapsin I (SYN1), regulated by TFs such as REST, was also markedly downregulated in glioma. Previous evidence indicates that SYN1 may promote tumor initiation by maintaining stemness, underscoring its potential as a therapeutic target at critical regulatory nodes [27, 28]. RAB3A, a small GTPase, has been reported to show context-dependent expression changes in glioma. Its role in tumorigenesis and drug resistance appears complex, highlighting its function as a molecular “switch” in malignant progression [29, 30]. SLC17A7 (VGLUT1), a glutamate transporter, was significantly downregulated in GBM, leading to disruption of neuron–glia neurotransmission, promotion of stemness, and facilitation of malignant progression, thus emphasizing its relevance for differentiation-based and anti-stemness therapeutic strategies [31, 32]. Synapsin II (SYN2), another synaptic vesicle regulator, may not only impair synaptogenesis and neurotransmission but also participate in neuron–tumor interactions and tumor microenvironment remodeling [33]. STXBP1 downregulation may influence ferroptosis and epithelial–mesenchymal transition (EMT), thereby attenuating proliferation, migration, and invasion, suggesting a tumor-suppressive role and potential for novel targeted therapy [34–37]. Overall, the downregulation of these neuron-related genes reflects a strategy by which glioma suppresses synaptic and neuronal functions while enhancing stemness and invasiveness, ultimately exacerbating malignancy. Pathway enrichment and immune infiltration analyses revealed that low expression of the core genes was not only associated with classical oncogenic processes such as cell cycle dysregulation and DNA replication, but also with activation of the complement and coagulation cascades, graft rejection, and other immune abnormalities. These results are consistent with the uniquely immunosuppressive microenvironment characteristic of glioma. Previous studies have demonstrated that macrophages and microglia play pivotal roles in promoting glioma immune evasion [38–40]. In this study, we further observed that downregulation of the core genes was negatively correlated with M1/M2 macrophages and NK cells, suggesting that their silencing may indirectly impair immune cell recruitment and activation, thereby contributing to the establishment of an immunosuppressive milieu. Single-cell analysis further revealed that the six core genes were predominantly localized within tumor-derived cellular subpopulations such as TPC-like and NPC-like cells, while exhibiting low expression in immune and other non-malignant cells in the tumor periphery. Notably, TPC-like cells displayed highly active malignant characteristics, with particularly strong expression of the core genes within their cluster region. This pattern suggests that these genes may primarily mark tumor stem–like cells or malignant populations endowed with synapse-like communication capabilities, consistent with recent findings describing “neuron-like tumor cells” in glioma [41]. These observations not only support, at the single-cell resolution, the concept of pseudo-neuronal transformation in tumor cells but also indicate that the core genes may serve as key molecular nodes driving this process. At the regulatory level, this study constructed a multilayer regulatory framework through ceRNA-based predictions and transcription factor activity inference using the decoupler-python platform. The ceRNA network suggested that multiple miRNAs and lncRNAs may regulate core gene expression through competitive interactions, whereas TF activity analysis highlighted TFs such as REST, SP1, ETV4, and EGR1 as possessing strong regulatory potential. These TFs are broadly implicated in neurodevelopment, synaptic plasticity, and tumorigenesis, suggesting that the core genes are subject to a complex interplay of transcriptional and post-transcriptional regulation [42–45]. When integrated with the distinct TF activity patterns observed in TPC-like cells, these findings suggest that the core genes are governed by specialized regulatory programs within malignant subpopulations—potentially promoting tumor-dominant behaviors associated with neuronal circuitry, water-channel dynamics, and synapse-like intercellular communication. Drug enrichment analysis further revealed significant associations between these genes and neurotransmitter-related or neuropsychiatric agents, including glutamate, GABA, and fluoxetine. These findings are consistent with emerging evidence that certain neuropsychiatric drugs may exert anti-tumor effects [46, 47], thereby providing novel perspectives for drug repurposing and individualized therapy in glioma. At the clinical application level, the six network hub genes consistently exhibited stable downregulation across multiple independent cohorts, and their ROC curve AUC values of approximately 0.8 indicate good diagnostic performance. Furthermore, experimental validation using 69 clinical glioma tissues and paired adjacent non-tumor samples reinforced the reliability of these genes as diagnostic and prognostic biomarkers. Although detailed survival and treatment information were not uniformly available across all public datasets, and a formal survival-based prognostic signature was therefore not constructed in this study, the six-gene-based nomogram and DCA support their potential clinical utility for glioma risk assessment. Future work using well-annotated survival cohorts will be required to comprehensively evaluate their prognostic value and to refine the model for outcome prediction. Taken together, this study elucidates the deep functional roles of the six network hub genes in glioma through a multi-dimensional framework encompassing bulk transcriptomics, single-cell analysis, immune ecosystem profiling, and regulatory network inference. Future work integrating spatial transcriptomics, organoid-based disease models, and CRISPR functional screening could further validate their biological functions and assess their clinical feasibility as therapeutic targets and candidates for drug repurposing. Conclusion By integrating five publicly available transcriptomic cohorts and systematically applying differential expression analysis together with WGCNA, this study identified 409 glioma-associated candidate key genes from a total of 504 samples. Through multi-algorithm network screening, including CytoHubba, LASSO regression, SVM-RFE, and random forest analysis, six consistently downregulated hub genes with stable diagnostic performance were ultimately identified: SYP, SYN1, RAB3A, SLC17A7, SYN2, and STXBP1. These genes were validated across independent cohorts and via qRT-PCR, and were further incorporated into a robust predictive model for glioma risk assessment. Collectively, these findings provide a solid foundation for understanding the molecular regulatory network of glioma and support the development of potential diagnostic biomarkers and targeted therapeutic strategies. Supplementary Information Supplementary Material 1. Supplementary Material 2. Figure S1. Workflow of weighted gene co-expression network analysis (WGCNA) in glioma. (a) Sample clustering dendrogram and trait heatmap. The x-axis represents individual samples, and the y-axis shows the clustering distance among samples. The heatmap below indicates sample grouping information (non-tumor vs. glioma). (b) Gene clustering dendrogram and initial dynamic tree cut modules. A weighted co-expression network was constructed using scale-free topology criteria. Initial co-expression modules were identified using the dynamic tree cut algorithm, with different colors representing distinct modules. (c) Merged dynamic modules after eigengene correlation–based merging. Modules with highly correlated eigengenes were merged to obtain biologically more meaningful final modules. (d) Clustering dendrogram of module eigengenes. Correlations among module eigengenes are visualized through the dendrogram, where closely related modules cluster together. (e) Distribution of gene significance (GS) across modules. The turquoise module exhibited the highest GS values, indicating its strongest association with the glioma trait. Supplementary Material 3. Figure S2. Protein–protein interaction (PPI) network of intersecting genes between DEGs and weighted gene co-expression network analysis (WGCNA) modules. The PPI network was constructed using the STRING database based on the intersecting genes from differentially expressed genes (DEGs) and WGCNA modules. Nodes represent proteins, edges indicate protein–protein interactions, and node colors denote different functional modules. Supplementary Material 4. Figure S3. Quality control and cell type identification of single-cell transcriptome data. (a) Enriched signaling pathways in the high RAB3A expression group. (b) Enriched signaling pathways in the high SLC17A7 expression group. (c) Enriched signaling pathways in the high STXBP1 expression group. (d) Enriched signaling pathways in the high SYN1 expression group. (e) Enriched signaling pathways in the high SYN2 expression group. (f) Enriched signaling pathways in the high SYP expression group. Supplementary Material 5. Figure S4. Quality control, clustering, and cell-type annotation of single-cell data. (a) Uniform Manifold Approximation and Projection (UMAP) visualization of single-cell quality-control metrics. (b) Cell clustering and annotation results of the integrated dataset. (c) Comparison of cell-type distributions between tumor core and tumor periphery regions. Supplementary Material 6. Figure S5. A regulatory network of network core genes constructed based on the competing endogenous RNA (ceRNA) hypothesis. This figure illustrates the potential ceRNA network of glioma network core genes (red circular nodes: SYP, SYN1, RAB3A, SLC17A7, and STXBP1). Green nodes represent microRNAs (miRNAs) predicted to interact with the core genes through the integration of TargetScan, miRDB, and miRanda databases, while blue nodes represent upstream long non-coding RNAs (lncRNAs) predicted to bind these miRNAs using the spongeScan database. Supplementary Material 7. Figure S6. Drug enrichment analysis of glioma candidate genes. Bubble plot showing drugs or small-molecule compounds significantly associated with the candidate genes. The x-axis represents the GeneRatio, bubble size indicates the number of associated genes, and bubble color reflects the adjusted p-value (p.adjust) after multiple testing correction. Supplementary Material 8. Table S1. Differentially expressed genes (DEGs) identified between glioma and normal tissues. Supplementary Material 9. Table S2. Candidate genes obtained from the intersection of differentially expressed genes (DEGs) and weighted gene co-expression network analysis (WGCNA) modules. Supplementary Material 10. Table S3. Potential hub genes identified by four CytoHubba algorithms (EPC, MCC, MNC, and Degree). Supplementary Material 11. Table S4. GSEA results for RAB3A based on high vs. low expression in glioma. Supplementary Material 12. Table S5. GSEA results for SLC17A7 based on high vs. low expression in glioma. Supplementary Material 13. Table S6. GSEA results for STXBP1 based on high vs. low expression in glioma. Supplementary Material 14. Table S7. GSEA results for SYN1 based on high vs. low expression in glioma. Supplementary Material 15. Table S8. GSEA results for SYN2 based on high vs. low expression in glioma. Supplementary Material 16. Table S9. GSEA results for SYP based on high vs. low expression in glioma. Acknowledgements Not applicable. Abbreviations Astrocyte-like Area under the curve Competing endogenous RNA Decision curve analysis Differentially expressed genes Disease Ontology Drug Signatures Database Endothelial cell Extracellular matrix Early growth response 1 Epidermal growth factor receptor Epithelial–mesenchymal transition Edge percolated component ETS variant transcription factor 4 Gamma-aminobutyric acid Glyceraldehyde-3-phosphate dehydrogenase Glioblastoma multiforme Gene Expression Omnibus Gene significance Gene set enrichment analysis Isocitrate dehydrogenase Kyoto Encyclopedia of Genes and Genomes Least absolute shrinkage and selection operator Long non-coding RNA Lymphocyte Macrophage Mast cell Maximal clique centrality Mesenchymal-like O6-methylguanine-DNA methyltransferase MicroRNA Maximum neighborhood component Natural killer N-methyl-D-aspartic acid Neural progenitor cell-like Nuclear respiratory factor 1 Oligodendrocyte Oligodendrocyte progenitor cell-like Principal component analysis Principal component Polymerase chain reaction Protein–protein interaction Quantitative real-time polymerase chain reaction Ras-related protein Rab-3 A RE1-silencing transcription factor Random forest Receiver operating characteristic Solute carrier family 17 member 7 Single-cell RNA sequencing Single-cell variational inference Standard deviation Specificity protein 1 Syntaxin binding protein 1 Support vector machine–recursive feature elimination Synapsin I Synapsin II Synaptophysin Transcription factor Telomerase reverse transcriptase Topological overlap matrix Tumor-propagating cell-like t-distributed stochastic neighbor embedding Uniform Manifold Approximation and Projection Unique molecular identifier Vesicular glutamate transporter 1 Weighted gene co-expression network analysis Authors’ contributions Xingxing Jiang, Yongping Gui and Yu Yang performed data acquisition, bioinformatic analysis, and drafted the manuscript. Xingxing Jiang assisted with data analysis and figure preparation. Aijun Liang designed and supervised the study and revised the manuscript. All authors read and approved the final manuscript. Funding This study was not supported by any external funding. Data availability The datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) under the accession numbers GSE109857, GSE15824, GSE35158, GSE4290, and GSE90886. Single-cell RNA sequencing datasets were also obtained from GEO, including GSE273274, GSE162631, and GSE182109. The clinical validation data generated during the current study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate All clinical samples used in this study were obtained with informed consent from patients and approved by the Ethics Committee of Jiangxi Provincial People’s Hospital (Approval No.: 2020-066). The study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Figure S1. Workflow of weighted gene co-expression network analysis (WGCNA) in glioma. (a) Sample clustering dendrogram and trait heatmap. The x-axis represents individual samples, and the y-axis shows the clustering distance among samples. The heatmap below indicates sample grouping information (non-tumor vs. glioma). (b) Gene clustering dendrogram and initial dynamic tree cut modules. A weighted co-expression network was constructed using scale-free topology criteria. Initial co-expression modules were identified using the dynamic tree cut algorithm, with different colors representing distinct modules. (c) Merged dynamic modules after eigengene correlation–based merging. Modules with highly correlated eigengenes were merged to obtain biologically more meaningful final modules. (d) Clustering dendrogram of module eigengenes. Correlations among module eigengenes are visualized through the dendrogram, where closely related modules cluster together. (e) Distribution of gene significance (GS) across modules. The turquoise module exhibited the highest GS values, indicating its strongest association with the glioma trait. Supplementary Material 3. Figure S2. Protein–protein interaction (PPI) network of intersecting genes between DEGs and weighted gene co-expression network analysis (WGCNA) modules. The PPI network was constructed using the STRING database based on the intersecting genes from differentially expressed genes (DEGs) and WGCNA modules. Nodes represent proteins, edges indicate protein–protein interactions, and node colors denote different functional modules. Supplementary Material 4. Figure S3. Quality control and cell type identification of single-cell transcriptome data. (a) Enriched signaling pathways in the high RAB3A expression group. (b) Enriched signaling pathways in the high SLC17A7 expression group. (c) Enriched signaling pathways in the high STXBP1 expression group. (d) Enriched signaling pathways in the high SYN1 expression group. (e) Enriched signaling pathways in the high SYN2 expression group. (f) Enriched signaling pathways in the high SYP expression group. Supplementary Material 5. Figure S4. Quality control, clustering, and cell-type annotation of single-cell data. (a) Uniform Manifold Approximation and Projection (UMAP) visualization of single-cell quality-control metrics. (b) Cell clustering and annotation results of the integrated dataset. (c) Comparison of cell-type distributions between tumor core and tumor periphery regions. Supplementary Material 6. Figure S5. A regulatory network of network core genes constructed based on the competing endogenous RNA (ceRNA) hypothesis. This figure illustrates the potential ceRNA network of glioma network core genes (red circular nodes: SYP, SYN1, RAB3A, SLC17A7, and STXBP1). Green nodes represent microRNAs (miRNAs) predicted to interact with the core genes through the integration of TargetScan, miRDB, and miRanda databases, while blue nodes represent upstream long non-coding RNAs (lncRNAs) predicted to bind these miRNAs using the spongeScan database. Supplementary Material 7. Figure S6. Drug enrichment analysis of glioma candidate genes. Bubble plot showing drugs or small-molecule compounds significantly associated with the candidate genes. The x-axis represents the GeneRatio, bubble size indicates the number of associated genes, and bubble color reflects the adjusted p-value (p.adjust) after multiple testing correction. Supplementary Material 8. Table S1. Differentially expressed genes (DEGs) identified between glioma and normal tissues. Supplementary Material 9. Table S2. Candidate genes obtained from the intersection of differentially expressed genes (DEGs) and weighted gene co-expression network analysis (WGCNA) modules. Supplementary Material 10. Table S3. Potential hub genes identified by four CytoHubba algorithms (EPC, MCC, MNC, and Degree). Supplementary Material 11. Table S4. GSEA results for RAB3A based on high vs. low expression in glioma. Supplementary Material 12. Table S5. GSEA results for SLC17A7 based on high vs. low expression in glioma. Supplementary Material 13. Table S6. GSEA results for STXBP1 based on high vs. low expression in glioma. Supplementary Material 14. Table S7. GSEA results for SYN1 based on high vs. low expression in glioma. Supplementary Material 15. Table S8. GSEA results for SYN2 based on high vs. low expression in glioma. Supplementary Material 16. Table S9. GSEA results for SYP based on high vs. low expression in glioma. Data Availability Statement The datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) under the accession numbers GSE109857, GSE15824, GSE35158, GSE4290, and GSE90886. Single-cell RNA sequencing datasets were also obtained from GEO, including GSE273274, GSE162631, and GSE182109. The clinical validation data generated during the current study are available from the corresponding author upon reasonable request.