1 Introduction

Primary liver cancer is the third most common cause of cancer-related mortality worldwide, with hepatocellular carcinoma (HCC) being the most prevalent type. It usually develops on a background of chronic liver disease and cirrhosis [1, 2]. Furthermore, HCC often presents at advanced stages, leading to poor patient prognosis. Given the continuously rising global HCC mortality rate, there is an urgent need for novel therapeutic approaches to address this significant clinical problem. Therefore, identifying key molecules in HCC and discovering early diagnostic markers and potential therapeutic targets are crucial for reducing the disease’s fatality rate [3].

The tumor microenvironment (TME) of HCC has been shown to be particularly immunosuppressive [4]. Cytotoxic lymphocytes within the microenvironment are suppressed, while a large number of leukocytes from various immune cell lineages participate in regulating lymphocyte activity [5]. Consequently, immune cells other than lymphocytes are now a focus of HCC research. As the first effector cells to arrive at sites of infection, inflammation, and tissue injury, neutrophils are key mediators of the innate inflammatory response and constitute a considerable proportion of the immune infiltrate in HCC [5, 6]. They exert antimicrobial and inflammatory functions, and participate in tumor progression through phagocytosis, degranulation, release of Neutrophil Extracellular Traps (NETs), and antigen presentation. NETs are web-like extracellular structures containing various antimicrobial components, many of which are directly related to altered cancer biology, including: neutrophil elastase (NE), matrix metalloproteinase 9 (MMP9), cathepsin G, histones, and DNA [7]. NETs play important roles in tumor progression, migration, invasion, forming an immunosuppressive niche, and promoting metastasis [8]. Although NETs have potential as therapeutic targets, drugs targeting NETs remain limited. To date, only two FDA-approved compounds can target NETs: inhaled DNase I and disulfiram [8,9,10]. However, they are not specific drugs for NETs, have limited clinical efficacy, and their use faces challenges such as delivery efficiency and potential side effects. Therefore, there is an urgent need to investigate NETs-related biomarkers to better understand HCC pathophysiology and develop more effective diagnostic and therapeutic strategies.

Advances in high-throughput sequencing and bioinformatics now enable comprehensive analysis of gene expression profiles to identify disease-specific molecular signatures. However, the specific role of NETs in HCC pathogenesis remains poorly defined. A systematic PubMed search (2000–2024) reveals fewer than 10 original studies directly examining NETs in human HCC tissues, highlighting a significant knowledge gap regarding their molecular mechanisms and clinical relevance. In this context, where functional and clinical validation data are scarce, an in‑silico approach provides a justified and efficient strategy to generate hypotheses and prioritize candidate biomarkers. This study therefore utilizes public transcriptomic datasets for an integrated bioinformatic analysis. By applying multiple machine learning algorithms, WGCNA, and immune infiltration analysis, we aim to identify and validate key NET‑related molecular biomarkers with clinical predictive power for HCC, while also proposing potential therapeutic targets through drug‑gene interaction analysis.

2 Materials and methods

2.1 Data download and preprocessing

The keyword “hepatocellular carcinoma” was used to search for HCC gene expression profiles in the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/GEO/). To ensure accuracy and reliability, the following criteria were applied for filtering: (1) HCC sequencing results must be derived from the analysis of human samples; (2) Datasets must include both treat and control groups, with each group containing more than six samples to ensure analytical robustness; (3) Datasets must have balanced data distribution; (4) The datasets with significant batch effects were excluded. According to these criteria, GEO datasets with accession numbers GSE60502, GSE76427, GSE113996, GSE121248, and GSE174570 were selected. Platform and grouping information for the datasets used in this study are provided in Supplementary file 1. After importing the data into R software (version 4.3.2, https://www.r-project.org/), probes were converted to gene symbols based on the platform annotation information of the normalized data. To maintain data integrity, probes without corresponding gene symbols were excluded. For probes mapped to the same gene, the average expression value was used as the final expression value to ensure accuracy and consistency. The “limma” package was used to merge the expression profiles, and the merged data was subsequently used for further analysis. NETs-related genes (NRGs) were sourced from two channels: those retrieved from the GeneCards database using the search term “Neutrophil Extracellular Trap” (https://www.genecards.org/Search/Keyord?queryString=%20Neutrophil%20Extracellular%20Trap&geneCategories=ProteinCoding) with a minimum relevance score of 10 were selected (supplementary file 2), and those identified through literature search (supplementary file 3).

2.2 Differentially expressed genes (DEGs) screening and functional enrichment analysis

The “limma” package was used to screen for DEGs between the treat and control groups in the merged dataset. The differential threshold was set at |log2FC| ≥ 0.585 (equivalent to a 1.5-fold change) and an adjusted p-value < 0.05. The “ggplot2” package was applied to create volcano plots and the “Heatmap” package to generate heatmaps for visualizing these DEGs. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed on the identified DEGs using the “org.Hs.eg.db”, “clusterProfiler”, “enrichplot”, and “ggplot2” packages, and the results were visualized using bubble plots.

2.3 Weighted gene co-expression network analysis (WGCNA)

WGCNA was used to identify co-expressed gene modules, study the relationships between gene networks and phenotypes, and identify hub genes within these networks. The “WGCNA” package was used to construct a co-expression network for all genes in the dataset. The topological overlap measure (TOM) was applied to identify all DEGs for further analysis. To estimate network connectivity, the weighted adjacency matrix was transformed into a TOM, and a hierarchical clustering method was used to construct its cluster tree structure. In the generated cluster tree, branches represent different gene modules, and colors represent different modules. Genes were classified based on their expression patterns, with genes sharing similar functions grouped into the same module. All genes were ultimately divided into multiple modules based on their weighted correlation coefficients. We used the pickSoftThreshold function from the WGCNA package to determine the optimal soft-thresholding power [11]. This function evaluates the scale-free topology fit index (R2) and mean connectivity (k) across a range of powers to determine the power that best approximates a scale-free network topology. The optimal power is typically chosen based on the highest R2 value and a relatively low mean connectivity, indicating a network with strong hubs and sparse connections.

2.4 Single-sample gene set enrichment analysis (ssGSEA)

The differential abundances of 28 immune-infltrating cells were computed using the ssGSEA algorithm by “GSVA” package. Immune cell abundances were visualized through a heatmap and violin plot, generated with R packages of “corrplot” and “ggplot2”. The correlations between DENRGs and immune cells were evaluated using Spearman’s correlation analysis with the R package “ggpubr”.

2.5 Machine learning for screening key biomarkers

To further screening key biomarkers associated with HCC and NETs, we used three machine learning algorithm including LASSO, SVM-RFE, and RF. LASSO was implemented using the R packages “glmnet” [12], To optimize the regularization parameter λ, cross-validation was performed with the cv.glmnet function. The search grid for λ was defined from 0 to 100, using the “binomial” family and the “class” metric (misclassification error) for evaluation. Subsequently, the final model was fitted using glmnet with the optimal λ (lambda.min), an alpha of 1 (L1 penalty), and the “binomial” family. SVM-RFE was implemented using the R packages“e1071” [13], parameter tuning was conducted via the tune.svm function, employing a linear kernel and testing cost values from 1 to 20. The final SVM model was constructed based on the optimal configuration identified through this process. RF was implemented using the R packages “randomForest” [14], we evaluated the error rate across tree counts ranging from 1 to 500. The optimal number of trees was determined via 10-fold cross-validation on the training set by selecting the configuration that yielded the lowest error. Additionally, feature importance scores were computed for each gene, with candidate hub genes defined as those having an importance value exceeding 5. The intersection of gene sets obtained from the three machine learning algorithms were considered as key biomarkers, and the results were visualized using the “Venn” package.

2.6 Validation of key biomarkers

The “pROC” package was further used to evaluate the diagnostic efficacy of key biomarkers. ROC curves were plotted, and an AUC value > 0.7 indicated high diagnostic potential. a nomogram model was established with “rms” package. Calibration curves and decision curve analysis (DCA) were applied to assess the model’s predictive accuracy. To validate our findings, we further used the external dataset GSE174570 to evaluate the expression of the key biomarkers and perform validation.

2.7 Unsupervised clustering and gene set variation analysis (GSVA)

The “ConsensusClusterPlus” package was used to perform consensus clustering on the expression profiles of key biomarkers in an unsupervised manner with 1000 iterations [15]. The cumulative distribution function (CDF) curve is a key indicator for assessing the stability of our consensus clustering analysis and determining the optimal number of subtypes. The k-means algorithm was used to cluster samples into different groups. The optimal cluster count was determined by evaluating the CDF curve and consensus matrix, ensuring a cluster consensus score above 0.9, with a maximum subtype range of 2 to 9 [15, 16]. Principal Component Analysis (PCA) was performed to highlight distinctions between subtypes. The “GSVA” and “GSEABase” packages were employed to identify variations in enriched gene sets between different clusters.

2.8 Prediction of potential drugs

Based on the identified diagnostic biomarkers, the DSigDB database (https://dsigdb.tanlab.org/DSigDBv1.0/download/html) was used to predict potential drugs for HCC treatment (Supplementary file 4). Cytoscape software (version 3.9.1) was used to visualize the biomarker-compound interaction network.

2.9 Statistical analysis

Unpaired Student’s t-test and Wilcoxon rank-sum test were used to analyze differences between two groups. Pearson or Spearman correlation analysis was used to assess relationships between variables. R software version 4.3.2 was used for statistical analysis and data visualization. Unless otherwise stated, a p-value < 0.05 was considered statistically significant.

3 Results

3.1 Identification and functional characterization of DEGs

The detailed workflow of the analysis is shown in Fig. 1. In the GSE60502, GSE76427, GSE113996, and GSE121248 datasets, we conducted a differential expression analysis for 223 HCC samples and 127 normal liver tissue samples. Transcriptomic analysis identified 826 DEGs, including 265 upregulated genes and 561 downregulated genes. Volcano plots and heatmaps effectively presented the significant findings (Fig. 2A, B).

Fig. 1
Fig. 1
Full size image

Flowchart of this study

Fig. 2
Fig. 2
Full size image

The volcano plot of DEGs between treat and control group The Heatmaps of DEGs between treat and control group The bubble plots of the GO enrichment analysis results for DEGs The bubble plot of KEGG pathway analysis results for DEGs

To explore the biological functions and potential roles of these DEGs in the occurrence and development of HCC, we performed GO and KEGG enrichment analyses. In the GO analysis (Fig. 2C), we found that in the Biological Process (BP) category, DEGs were mainly enriched in processes such as small molecule catabolic process, cellular catabolic process, organic acid catabolic process, and carboxylic acid catabolic process. In the Cellular Component (CC) category, DEGs were significantly enriched in collagen-containing extracellular matrix, secretory granule lumen, and cytoplasmic vesicle lumen. In the Molecular Function (MF) category, DEGs were primarily involved in important functions such as oxidoreductase activity, acting on paired donors, with incorporation or reduction of molecular oxygen, peptidase inhibitor activity, and heme binding.

KEGG pathway analysis (Fig. 2D) indicated that these DEGs were significantly enriched in various pathways, including Complement and coagulation cascades, Cell cycle, Carbon metabolism, Chemical carcinogenesis - DNA adducts, Biosynthesis of amino acids, and Metabolism of xenobiotics by cytochrome P450 etc.

3.2 Gene module identification and co-expression network formation

The results of the WGCNA indicated that a soft threshold power of 20 was optimal (Fig. 3A and B). This led to the generation of a merged cluster tree module (Fig. 3C) and subsequent module-trait relationship assessment (Fig. 3D), which identified the brown module with the highest correlation coefficient, containing 362 genes.

Fig. 3
Fig. 3
Full size image

A clustered dendrogram between treat and control group The soft threshold analysis diagram The clustering tree module diagram Heatmaps showing module–trait relationships based on WGCNA

3.3 Identifying differentially expressed NET-related genes (DENRGs)

The NRGs was retrieved from the GeneCards website, containing 627 genes. By taking the intersection with the 362 genes from WGCNA and the 826 DEGs, a total of 18 key DENRGs were screened (Fig. 4A). Box plots showed significant expression differences of the 18 DENRGs between the treat and control groups. Compared to the control group, 2 genes (PSMB4, ITGA6) showed statistically significant upregulation in treat group, while the rest were significantly downregulated (ECM1, DNASE1L3, CD4, MAP2K1, FOS, IL18R1, FCGR2B, TEK, EGR1, JUN, FCGR3B, A2M, IL33, SERPINE1, PTGS2, CRP) (Fig. 4B). Corresponding heatmaps and Spearman correlation analysis displayed the correlations among the intersected genes (Fig. 4C). A circular diagram visualized the distribution of DENRGs on chromosomes (Fig. 4D). A Manhattan plot illustrated the differential expression of DENRGs between the treat and control groups and their distribution on chromosomes (Fig. 4E).

Fig. 4
Fig. 4
Full size image

Venn diagram of DENRGs Histogram of the expression of 18 DENRGs between treat and control group Spearman’s correlation analysis of 18 DENRGs Chromosomal positions of the 18 DENRGs Manhattan plot illustrated the differential expression of DENRGs between the treat and control groups and their distribution on chromosomes Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent standard deviation

3.4 Immune cell infiltration in HCC

The Heatmap revealed significant differences of 28 immune cell types between treat and control groups (Fig. 5A). Notably, the algorithm found that, compared to the control group, the proportions of most immune cells were significantly reduced in treat groups (p < 0.001), except for Activated CD4 T cell (p = 0.342), CD56 bright natural killer cell (p = 0.808), and Effector memory CD8 T cell (p = 0.071) (Fig. 5B). Furthermore, correlation analysis showed that the 18 DENRGs were associated with Type 1 T helper cell, Regulatory T cell, Neutrophil, Macrophage etc. (Fig. 5C).

Fig. 5
Fig. 5
Full size image

The heatmap exhibited the difference of 28 types of immune cells between treat and control group A violin graph illustrated the difference of immune cells between treat and control group Correlation between DENRGs and immune cells Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent standard deviation

3.5 Identification of key biomarkers in HCC

LASSO, SVM-RFE, and RF algorithms were applied to identify reliable diagnostic biomarkers associated with HCC from these DENRGs. The LASSO algorithm identified 9 key genes and determined the optimal lambda value (Fig. 6A, B). SVM-RFE analysis obtained 12 key genes as reliable molecular biomarkers for HCC (Fig. 6C, D). The RF algorithm effectively screened 11 important genes (Fig. 6E, F). The intersection of results from the three machine learning methods identified three key biomarkers in HCC: ECM1, DNASE1L3, JUN (Fig. 6G). All three biomarkers were found to be downregulated in treat group (Fig. 6H).

Fig. 6
Fig. 6
Full size image

(A, B) The LASSO algorithm identified 9 key genes and determined the optimal lambda value (C, D) SVM-RFE analysis obtained 12 key genes (E, F) The error rate confidence intervals and the relative importance of genes for RF algorithm (G) The Venn diagram of the intersection of LASSO, SVM-REF, and RF signature genes (H) Histogram of the expression of three key biomarkers between treat and control group Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent standard deviation

3.6 Validation of the diagnostic value of key biomarkers

To evaluate the diagnostic effectiveness of the three biomarkers, we plotted ROC curves. The results showed that ECM1 had an AUC of 0.920; DNASE1L3 had an AUC of 0.881; JUN had an AUC of 0.814; and the combined AUC was 0.924 (Fig. 7A, B). Then, a nomogram model was constructed based on the three genes. The relative expression level of each gene corresponded to a score, and the total score was calculated by adding up the scores of each gene. Higher expression levels of ECM1 corresponded to higher diagnostic scores for HCC, indicating that ECM1 expression level is crucial in HCC diagnosis (Fig. 7C). The calibration curve showed a moderate difference between the predicted and actual cluster risk for HCC (Fig. 7D). DCA results confirmed the high accuracy of the nomogram (Fig. 7E). Simultaneously, we also applied the external validation set GSE174570 to further evaluated. Differential analysis results showed that compared to the control group, all three biomarkers were significantly downregulated in the treat group (Fig. 7F). ROC analysis results showed that ECM1 had an AUC of 0.975; DNASE1L3 had an AUC of 0.980; JUN had an AUC of 0.790; and the combined AUC was 0.986 (Fig. 7G, H). The results indicate that the three markers have good diagnostic capability.

Fig. 7
Fig. 7
Full size image

ROC curves of the three key biomarkers alone for diagnosis. ROC curves of the three key biomarkers combination for diagnosis. The nomogram established based on three key biomarkers. The calibration curve of the nomogram. The DCA results confirmed the accuracy of the nomogram. Histogram of the expression of three key biomarkers between treat and control group in the external validation set. ROC curves of the three key biomarkers alone for diagnosis in the external validation set. ROC curves of the three key biomarkers combination for diagnosis in the external validation set. Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent standard deviation

3.7 Determination of key biomarkers molecular clusters in HCC

Consensus clustering analysis showed that HCC patients were robustly divided into two distinct molecular subtypes (Fig. 8A). As shown in Fig. 8B, the CDF curve showed significant improvement when moving from k = 2 (red line) to higher clusters, with minimal additional gain beyond k = 3, demonstrating the stability of our two-cluster model. This optimal clustering was visualized in the consensus matrix, showing strong within-cluster consensus and clear separation between clusters. The PCA plot further validated this classification, with samples from clusters C1 and C2 forming distinct groups in the dimensional space (Fig. 8C). Histograms showed that the biomarkers exhibited low expression levels in the C1 cluster (Fig. 8D). Subsequently, we examined the distinction in immune microenvironment characteristics between the two clusters and found that the C1 cluster shared the same immune microenvironment as the overall HCC population, representing the typical disease subtype (Fig. 8E). Functional enrichment analysis indicated that the C2 cluster compared to cluster C1 showed upregulation in pathway such as complement and coagulation cascades, fatty acid metabolism, and bile acid biosynthesis etc. (Fig. 8F).

Fig. 8
Fig. 8
Full size image

Consensus clustering analysis showed two clusters (k = 2). the CDF curve demonstrated the stability of two cluster model. PCA analysis of the two clusters. Histogram of three key biomarkers expression levels between two clusters. Different expression levels of immune cells between cluster 1 and cluster 2. GSVA analysis between cluster 1 and cluster 2. *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent standard deviation

3.8 Drug-gene interactions

Two genes were targeted by 16 drugs for inhibition. Furthermore, Cytoscape software was used to visualize the drug-gene interaction network (Fig. 9).

Fig. 9
Fig. 9
Full size image

Drugs-genes network of two key biomarkers

4 Discussion

In this study, we employed an integrated bioinformatics approach to investigate NET-related molecular features in HCC and identified key biomarkers with potential diagnostic value. Differential analysis results revealed 826 differentially expressed genes. To explore the biological functions and potential roles of these DEGs in the occurrence and development of HCC, we performed GO and KEGG enrichment analyses. DEGs were predominantly enriched in GO and KEGG pathways associated with catabolic process, collagen-containing extracellular matrix, cytoplasmic vesicle lumen, oxidoreductase activity, cell cycle etc. The process of NET formation requires catabolic processes under oxidoreductase activity to remove dense chromatin and the actin cytoskeleton, leading to lytic cell death of neutrophils [8]. This provides favorable evidence for further exploration of molecular mechanisms.

A key finding of our study was the identification of 18 DENRGs through the intersection of DEGs, co-expression modules, and known NET-related genes. In our analysis, 16 DENRGs were significantly downregulated in treat group compared to the control group. ssGSEA results showed that most immune cells were significantly downregulated in HCC, further indicating that HCC exists in an immunosuppressive microenvironment during its development. Various immunosuppressive cells are believed to drive tumor growth and metastasis and are significantly associated with the prognosis of hepatocellular carcinoma patients, including MDSCs, Tregs, and macrophages, etc [4, 17]. Our correlation analysis results showed significant correlations between the 18 DENRGs and various immune cells such as neutrophils, MDSCs, macrophages, and helper cells. DENRGs play a key role in regulating immune cell infiltration and molecular processes, but how NETs participate in regulating this immune microenvironment requires further study.

By applying machine learning algorithm, we screened three key biomarkers (ECM1, DNASE1L3, JUN) for HCC diagnosis. ECM1 is a secreted protein that plays a crucial role in the extracellular matrix, influencing cell behavior, tissue remodeling, and immune response regulation [18]. ECM1 initially discovered in the epidermis and dermis, its function is not only to maintain skin integrity and homeostasis, but also involved in cancer development, such as in prostate cancer, breast cancer, gastric cancer, and colorectal cancer, etc [19,20,21,22]. Recent studies indicate that ECM1 plays an important role in fatty liver disease, particularly in the development of liver fibrosis [23]. In this study, ECM1 expression was lower in the treat group compared to the control group. Liu et al. [24] also screened ECM1 as one of the main diagnostic markers using machine learning methods, and ECM1 expression was similarly lower in the HCC group. However, few studies reported the mechanism of ECM1 in HCC. Deoxyribonuclease 1 Like 3 (DNASE1L3) belongs to the DNase 1 family and is responsible for cleaving chromatin DNA in an internucleosomal manner, independent of proteolytic assistance, under physiological conditions, DNASE1L3 is highly expressed by dendritic cells (DCs) and macrophages in the spleen and liver [25]. Several studies have demonstrated the tumor-suppressive role of DNASE1L3 in HCC, high expression of DNASE1L3 can inhibit HCC progression through various pathways [25,26,27]. Wang et al. [26] showed that DNASE1L3-induced PANoptosis enhances the activation of anti-tumor immunity in the tumor microenvironment (TME), thereby improving the efficacy of combined sorafenib and PD-1 monoclonal antibody treatment. Chen et al. [25] found that dendritic cell-derived DNASE1L3 can mediate the degradation of NETs, promoting anti-tumor immune responses and enhancing anti-PD-(L)1 efficacy. Consistent with the prediction results of this study, DNASE1L3 expression was lower in the HCC group, acting as a tumor suppressor gene. Jun proteins (c-Jun, JunB, JunD) are essential components of the dimeric AP-1 transcription factor complex, regulated by various signals such as growth factors, inflammatory cytokines, mechanical and oxidative stress, and play important roles in many diseases including cancer [28,29,30]. Among them, c-Jun plays a significant role in the pathogenesis and progression of hepatocellular carcinoma. Zhu et al. [30]found that c-Jun is a direct target gene of ZBTB7B and plays a central signaling node role in the development of ZBTB7B-deficient HCC. Bakiri L et al. [29]established a c- Jun ~ Fra- 2hep mouse model for HCC research and revealed that the expression of c-Jun/Fra-2 dimers in the liver can induce spontaneous tumors with HCC characteristics. Fibronectin also can promote the progression and metastasis of hepatocellular carcinoma through the JUN pathway [31]. Our findings suggest that NET-related genes regulate the immune microenvironment, and that NETs may promote tumor vascular invasion and metastasis by degrading endothelial tight junction molecules. This process contributes to the formation of large vascular tumor thrombi, which is closely associated with an immunosuppressive microenvironment and pathological angiogenesis. Patients with such thrombi (in the hepatic vein, inferior vena cava, and/or right atrium) face an extremely poor prognosis and limited therapeutic options, representing a major clinical challenge [2]. Notably, Chen QY et al. [32]reported that low expression of DNASE1L3 is significantly associated with tumor thrombus formation. Taken together, these insights provide a molecular rationale for exploring combined immunotherapy and anti-NETs therapy as a potential strategy for high-risk HCC subtypes.

In this study, the ROC curves and the nomogram model validated that the three markers have good diagnostic capability. The external validation set further confirmed the important value of these three biomarkers in diagnosing and predicting HCC. NETs have been shown to influence prognosis by engaging in biological processes such as thrombosis and angiogenesis, which directly modulate tumor blood supply [33]. Separately, imaging biomarkers like the Portal Venous Coefficient (PVC) and Hepatic Arterial Coefficient (HAC) have been reported by Ke, Y et al. as valuable independent predictors of postoperative survival in HCC patients [34]. However, it remains unclear whether and how NETs activity mechanistically links to these imaging vascular parameters (PVC/HAC). Therefore, further investigation is warranted to determine whether the three key biomarkers identified in our study (ECM1, DNASE1L3, JUN) influence PVC and HAC through NET-related pathways. If a future mechanistic link is established, it would provide a strong rationale for developing an integrated prognostic model that combines these molecular biomarkers with imaging-derived vascular parameters to improve postoperative survival prediction in HCC.

Consensus clustering is categorized as an unsupervised algorithm designed to discern potential groupings by analyzing the inherent characteristics of the data. This method identified 2 distinct molecular subtypes within the HCC dataset. Cluster 1 was identified as the typical disease subtype, characterized by the lower expression of key biomarkers and downregulation of immune cells. Furthermore, our drug-gene interaction analysis revealed potential therapeutic targets. However, only 2 genes were associated with drugs; JUN could be inhibited by 10 drugs, DNASE1L3 by 6 drugs, and no related inhibitors were found for ECM1.

However, this study has several limitations. First, although our findings are statistically robust, they are primarily based on transcriptomic data and require further functional validation. Second, there is limited research on the mechanisms of the key NETs genes, and their regulatory mechanisms need further clarification. Third, the potential therapeutic implications of targeting the identified biomarkers need to be evaluated in appropriate experimental models.

5 Conclusion

In summary, this study identified key genes associated with NETs that are involved in HCC progression, elucidating their roles in immune dysregulation. The clinical predictive value of three biomarkers (ECM1, DNASE1L3, JUN) for HCC was validated, and drug-gene interaction analysis revealed potential targets associated with these key genes. These findings enhance our understanding of the pathogenesis of HCC and provide promising opportunities for precise diagnosis and targeted therapy.