Integrating WGCNA and machine learning algorithm to identify ACSM5 as a prognostic biomarker and therapeutic target for predicting immunotherapy efficacy in non-small cell lung cancer

Introduction

Lung cancer is the most prevalent form of cancer in China, with non-small cell lung cancer (NSCLC) accounting for 85% of all cases (1). Surgical intervention is the primary treatment modality in the early stages of NSCLC (2). whereas patients with advanced-stage disease typically undergo chemotherapy, radiotherapy, targeted therapy, and immunotherapy (3,4). Chemotherapy and radiotherapy have major negative effects on patients (5-7), and targeted therapy necessitates specific mutational targets and is prone to the development of drug resistance (8). On the other hand, immunotherapy targets both primary and metastatic tumors by activating the systemic immune system (9), resulting in comprehensive therapeutic effects with minimal adverse responses and long-term efficacy. However, tumor heterogeneity significantly contributes to variability in patient response to immunotherapy (10-14). Consequently, there is an urgent need for improved technologies and methodologies to address these clinical challenges.

Despite the promise of immunotherapy in providing renewed hope for patients with advanced-stage NSCLC, the 5-year survival rate remains approximately 20% (15). Moreover, there are notable differences in therapeutic effectiveness between individuals (16,17). By thoroughly analyzing genes related to immunotherapy response, it is possible to accurately predict the likelihood of patients benefiting from treatment, thereby preventing the use of costly and ineffective medications in those who are unlikely to respond. This study aimed to elucidate the mechanisms by which tumor cells evade the immune system, providing a theoretical foundation for the development of novel immunotherapeutic strategies and promoting innovation in the field of immunotherapy (18). Weighted gene co-expression network analysis (WGCNA) organizes genes into functional modules based on their expression similarity and identifies key genes and modules associated with specific phenotypes or biological processes (19). This approach utilizes a weighted network model to represent gene co-expression relationships more accurately, enabling the investigation of gene interactions and their potential functions from a comprehensive perspective. Previous studies have applied the WGCNA method in the investigation of various cancers, including ovarian cancer (20), liver cancer (21) gastric cancer (22). In addition, the WGCNA strategy has been used to detect features related to radiation resistance (23). In this study, we initiated our investigation by applying WGCNA to sequencing data from GSE207422, a NSCLC dataset in the Gene Expression Omnibus (GEO), to identify genes associated with immunotherapy response.

Machine learning algorithms are widely employed in disease signature construction, prediction, and diagnosis (24-26), enhancing both the accuracy and efficiency of diagnoses and providing robust support for personalized and precision medicine (27-29). Beyond solid tumors, recent studies in hematologic malignancies provide compelling evidence: systematically evaluated machine learning applications in chronic myeloid leukemia (CML) (29). In this study, we conducted machine learning analysis based on the immunotherapy response genes identified from The Cancer Genome Atlas-lung adenocarcinoma (TCGA-LUAD) cohort and GEO dataset, and selected the algorithm with the highest c-index score. The prognostic capability of this signature was rigorously validated across various clinical subgroups and GEO cohorts, and it emerged as an independent prognostic factor through multivariable analysis. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1620/rc).

Methods

Data collection and processing

The GEO repository provided the accession code GSE207422, comprising a cohort of 24 samples obtained from patients with NSCLC, of which nine responded to immunotherapy and 15 did not. For our analysis, we retrieved 503 cases of lung adenocarcinoma (LUAD) and 54 normal cases from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/). The TCGA dataset was utilized as the training set, whereas the GEO datasets served as validation sets to enhance the robustness of our data analysis. The GEO datasets included GSE31210, GSE30219, and GSE29016. Detailed information on these datasets is provided in Table S1.

Immunotherapy-responsive genes (IRGs) availability

We utilized WGCNA to identify genes associated with immunotherapy responses. The WGCNA package (version 1.73) of the R software was used to analyze the genes in GSE207422 and construct a weighted gene co-expression network. An optimal soft threshold of 7 was chosen to obtain a scale-free network. A clustering tree diagram was then constructed based on the calculated gene adjacency and similarity, with the requirement that the number of genes in a module should be greater than or equal to 60. The dynamic tree cutting algorithm was utilized to segment the modules further and merge similar modules. The association between each gene module and sample traits was evaluated using Pearson’s correlation, and several modules with higher absolute values were selected for subsequent analysis. The genes within these modules were found to be related to immunotherapy response and are detailed in table(s) available at https://cdn.amegroups.cn/static/public/tcr-2025-1620-1.xls of the supplementary file.

Construct the prognostic mode

A total of 3,516 IRGs were identified. Subsequently, a differential analysis was performed on these genes within the TCGA dataset, employing a logarithmic fold change (logFC) threshold of 0.585 and a false discovery rate (FDR) threshold of 0.05. Following this, a Cox regression analysis was conducted on the selected genes, with the criterion for inclusion in the survival analysis being a P value from the Cox regression analysis (coxPfilter) not exceeding 0.05. Among these, 94 genes linked to prognosis were selected through univariate regression analysis, and are presented in Table S2.

For signature construction, a combination of 101 machine-learning algorithms (listed in Table S3) was utilized (30). including support vector machine (SVM), least absolute shrinkage and selection operator (LASSO), gradient boosting machine (GBM), Random Forest, Elastic Net, Stepwise Cox, Ridge, CoxBoost, Super Partial Correlation (SuperPC), and Partial Least Squares with Cox regression (plsRcox) (31). Parameter configurations are established through an integration of dynamic parsing, automated optimization via cross-validation, and the application of predetermined default values. TCGA dataset served as the training set, whereas GEO datasets were used for validation. We employed 10-fold stratified cross-validation using the “glmnet” R package. The model with the highest average C-index in all cohorts was deemed to be optimal. Based on this algorithm, 11 genes were selected for subsequent modeling.

The performance of the prognostic signatures

The “Survival” R package was used to perform survival analysis (32). Survival curves were generated using the “survminer” R package, which facilitates drawing survival curves utilizing the ‘ggplot2’ visualization framework. The “timeROC” R package was used to construct the receiver operating characteristic (ROC) curves (33). Subsequently, a predictive nomogram was developed using the “regplot” R package, incorporating the risk score along with other clinical variables to forecast clinical outcomes for NSCLC patients.

Evaluation of the immune microenvironment

The assessment of immunological scores, Stromal Scores, and ESTIMATE scores for the LUAD patient cohort was executed via the utilization of the “estimate” R package. To quantify immune-related functions in individual samples, we conducted a single-sample gene set enrichment analysis (ssGSEA) utilizing the “GSVA” and “GSEAbase” R packages. Additionally, we analyzed the risk score and established immune checkpoint genes (ICGs) within the TCGA-LUAD cohort. The TIMER2.0 database offers a comprehensive overview of immune cell infiltration levels in TCGA, while the IOBR program encompasses various immune infiltration methods used to determine the number of immune cells in tumor samples.

Mutational landscape and immunotherapeutic approaches

The “Maftools” package in R software was utilized to analyze mutated genes in NSCLC. This package enables the visualization and summarization of these genes through waterfall plots, providing a clear and concise representation of the mutational landscape. Additionally, the correlation between the risk score and tumor mutational burden (TMB) was analyzed and visualized using the “ggpubr” and “reshape2” R packages. The Cancer Immunome Atlas (available at https://tcia.at/home) was used to assess the IPS of NSCLC patients, identifying populations that may be suitable for immunotherapy. Furthermore, the “ggpubr” package in R was used to compare IPS and Tumor Immune Dysfunction and Exclusion (TIDE) scores, which are essential metrics for evaluating the therapeutic benefits of immunotherapy.

Quantitative reverse transcription polymerase chain reaction (qRT-PCR) assay

Clinical samples were collected from Tianjin Cancer Hospital, which were approved by the Ethical Board of the Tianjin Medical University Cancer Institute and Hospital (No. bc2023152), and informed consent was obtained from all participating patients. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Tissues were extracted from eight patients diagnosed with NSCLC using TRIZOL reagent (Qiagen, Carlsbad, CA, USA). Data analysis was conducted using the ΔΔCt method, with Gapdh serving as the internal reference gene. The experiment was repeated three times to ensure reliability and accuracy. Primers designed to amplify Acyl-CoA synthetase medium chain family member 5 (ACSM5) are provided below.

Forward Primer: GGACAGGGACTGTGATGATTCC;Reverse Primer: CCCTTGGAGCTAGGGAGTCA.

Statistical analysis

All statistical analyses were performed using R (v4.3.1) for data processing and visualization, and GraphPad Prism (v9.1.1) for statistical modeling. Experimental data, derived from three independent trials, are presented as mean ± standard deviation (SD). For parametric comparisons, a two-tailed Student’s t-test was used for two-group comparisons. Statistical significance (P≤0.05) is marked with asterisks, and “ns” indicates non-significance (P>0.05).

Results

Identification of immunotherapy-response genes via WGCNA analysis

The procedure of our study is illustrated in Figure 1.

Figure 1 The procedure of our study. PCR, polymerase chain reaction; TME, tumor microenvironment; WGCNA, weighted gene co-expression network analysis.

The “WGCNA” package in R was utilized to identify key modules, and the 24 samples from GSE207422 in the GEO database were categorized into two groups: Non-major pathological response (NMPR) and major pathological response (MPR). A soft threshold of eight was selected (Figure 2A). Subsequently, dynamic module identification was conducted across different cohorts, ensuring that each module contained no fewer than 60 genes (Figure 2B). A total of 21 co-expression modules were clustered, with the yellow and pink modules exhibiting the strongest positive correlations with immunotherapy-response genes (Cor =0.48, P=0.02) and (Cor =0.44, P=0.03), respectively (Figure 2C). To assess the significance of the genes, we calculated the correlation between the genes and phenotype, yielding gene significance (GS) values (Figure 2D,2E).

Figure 2 Identification of genes responsive to immunotherapy through WGCNA analysis and differential expression profiling. (A) Soft threshold power set at 7. (B) Genes with similar expression patterns clustered into co-expression modules in GSE207422. (C) Correlation between module eigengenes and phenotype. (D) MM vs. GS scatter plot (pink module). (E) MM vs. GS scatter plot (yellow module). (F) Heatmap of module gene differential analysis in TCGA data. (G) Volcano plot of module gene differential analysis in TCGA data. FC, fold change; GS, gene significance; MM, membership correlation; TCGA, The Cancer Genome Atlas; WGCNA, weighted gene co-expression network analysis.

Preparations prior to constructed signatures

We integrated immune response genes into the expression matrix data from TCGA and conducted differential analysis of IRGs within the TCGA dataset. The results are presented using a volcano plot and heatmap in Figure 2F,2G, respectively. To ensure comparability among the datasets, we mitigated batch effects across the three NSCLC transcriptome cohorts. Figure 3A displays the data prior to batch effect reduction, whereas Figure 3B illustrates the data after batch effect removal. Subsequently, Cox proportional hazards analysis was performed on these differentially expressed genes, identifying 94 genes associated with prognosis (Figure 3C).

Figure 3 Preparations prior to constructed signatures and constructed signatures. (A) The data prior to batch effect reduction. (B) The data after batch effect removal. (C) A Cox proportional hazards analysis. (D) Immunotherapy prognostic signatures using ML combinations; heatmap values show C-index for predicting prognosis. ML, machine learning.

Establishment and validation of the prognostic signatures

We employed a machine learning combinatorial algorithm to analyze the 94 prognostic genes, utilizing TCGA dataset as the training cohort and three GEO datasets as validation cohorts. Model selection was based on the average C-index across all four cohorts, ultimately identifying the CoxBoost + Ridge algorithm as the optimal model (Figure 3D). The genes of each algorithm combination and the selected model are listed in table(s) available at https://cdn.amegroups.cn/static/public/tcr-2025-1620-2.xls. Subsequently, patients were stratified into high-risk and low-risk groups based on the median risk score. Notably, across TCGA dataset (Figure 4A), GSE29016 (Figure 4B), GSE30219 (Figure 4C), and GSE31210 (Figure 4D), the high-risk group consistently exhibited poorer prognosis outcomes, thus validating the accuracy and robustness of our developed model. Furthermore, we assessed the potential of this risk model as an independent prognostic factor for NSCLC patients with NSCLC. Both univariate and multivariate Cox regression analyses revealed significant correlations between the risk score, disease stage, and overall survival (OS) (Figure 4E,4F). The ROC curves for the risk score, age, sex, and stage demonstrated AUC values of 0.737, 0.520, 0.600, and 0.730, respectively (Figure 4G). The ROC curves corresponding to the 1-, 3-, and 5-year time points derived from this signature for TCGA, GSE29016, GSE30219, and GSE31210 datasets are presented in Figure 4H-4K, respectively. To enhance the clinical utility of this risk model, we developed a nomogram to predict the 1-, 3-, and 5-year survival outcomes of patients with TCGA-NSCLC. As shown in the nomogram (Figure 5A), the risk score model exhibited the highest weighting among all the included variables, such as gender, age, and stage.

Figure 4 Validation of the prognostic signatures. (A-D) Survival analysis of high-risk and low-risk groups in (A) TCGA, (B) GSE29016, (C) GSE30219, and (D) GSE31210. (E) Multivariate Cox regression analyses. (F) Univariate Cox regression analyses. (G) ROC curves for risk score, age, gender, and stage, with corresponding AUC values displayed. The ROC curves corresponding to the 1-, 3-, and 5-year time points, derived from this signature for the (H) TCGA, (I) GSE29016, (J) GSE30219, and (K) GSE31210. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Figure 5 Construction of a prognostic nomogram and immune microenvironment characterization. (A) A nomogram to predict 1-, 3-, and 5-year survival outcomes in TCGA patients. (B) Violin plot showing differences in Stromal Score, Immune Score, and ESTIMATE Score (calculated using the ESTIMATE algorithm) between the two risk subgroups. (C) A box plot showed that differences in immune checkpoint gene expression between high- and low-risk groups. (D) Box plot showed that differences in immune function between high- and low-risk groups. *, P<0.05; **, P<0.01; ***, P<0.001. ESTIMATE, Estimation of STromal and Immune cells in MAlignant Tumor tissues using Expression; TCGA, The Cancer Genome Atlas; TME, tumor microenvironment.

Assessment of immune microenvironment

The ESTIMATE method was used to evaluate immune infiltration levels across different risk groups, revealing that the low-risk cohort exhibited significantly higher Immune Scores (P<0.05) (Figure 5B; tables available https://cdn.amegroups.cn/static/public/tcr-2025-1620-3.xls). Figure 5C depicts the correlation between established ICGs and the risk score in the TCGA-LUAD cohort. Notably, the low-risk group demonstrated higher expression levels of several immune checkpoint genes including TNFSF4, CD27, CD48, CD40LG, and CTLA-4. The distribution of immune cells, such as B cells, CD8⁺ T cells, macrophages, and neutrophils, as well as biological processes, such as cytolytic activity, T cell co-stimulation, and APC co-stimulation, indicate a more active immune function in the low-risk group across both risk categories, as illustrated in Figure 5D.

Analysis of TMB and immunotherapy in high- and low-risk groups

A comprehensive analysis of mutation frequency within the model gene was conducted. Figure 6A illustrates that genes such as MUC2, PZP, and ACSM5 exhibited a significantly higher mutation incidence. As depicted in Figure 6B, the TMB is notably lower in the low-risk group. Furthermore, a detailed examination of Figure 6C reveals a discernible positive correlation between TMB and risk score, suggesting a potential mechanistic link between these two variables within the studied biological system. A comprehensive analysis of the IPS and TIDE scores was conducted to clarify the potential therapeutic significance of immunotherapy across various risk stratifications. Individuals in the low-risk group were more likely to benefit from immunotherapy, as shown in Figure 6D-6H.

Figure 6 Mutation landscape and the performance of immunotherapy in high and low risk groups. (A) Comparison of mutations between high and low risk groups. (B) The TMB is notably lower in the low-risk group. (C) Positive correlation between TMB and the risk score. (D) TIDE scores in high-risk and low-risk groups. (E-H) Prediction of IPS for TCGA patients using The Cancer Immunome Atlas. *, P<0.05; **, P<0.01; ***, P<0.001. IPS, immune checkpoint inhibitor response prediction scores; TCGA, The Cancer Genome Atlas; TIDE, Tumor Immune Dysfunction and Exclusion; TMB, tumor mutational burden.

The role of ACSM5

For subsequent analysis, ACSM5 was selected from the model genes. ACSM5 exhibits high expression levels in normal tissues associated with NSCLC, whereas its expression is significantly reduced in tumor tissues, as illustrated in Figure 7A. The survival analysis curve demonstrated that NSCLC patients with high ACSM5 expression exhibited improved prognostic outcomes (Figure 7B). We conducted an experimental assessment of ACSM5 expression in normal and tumor tissues derived from patients with NSCLC. As illustrated in Figure 7C, there was a marked reduction in ACSM5 expression within the tumor tissues, which supports our previous analyses. Figure 7D,7E depict the association between ACSM5 expression in tumors and immune cell infiltration. Furthermore, analysis using Timer 2.0 indicates that ACSM5 expression was inversely correlated with tumor infiltration, but exhibited a positive correlation with CD4+, CD8+ T cells, and DC cells, as demonstrated in Figure 7E. These findings offer a more comprehensive understanding of the role of ACSM5 as a tumor suppressor within the tumor microenvironment.

Figure 7 The role of ACSM5. (A) The expression of ACSM5 is significantly reduced in tumor tissues. (B) The survival analysis curve in high and low level of ACSM5. (C) PCR assay of clinical samples. ACSM5 was highly expressed in NSCLC. (D) Relationship between ACSM5 and immune cell infiltration. (E) Timer 2.0 analysis reveals an inverse correlation between ACSM5 expression and tumor infiltration. *, P<0.05; **, P<0.01; ***, P<0.001. ACSM5, Acyl-CoA synthetase medium chain family member 5; NSCLC, non-small cell lung cancer; PCR, polymerase chain reaction; TPM, transcripts per million.

Discussion

Immune checkpoint blockade (ICB) therapy represents a critical treatment modality for advanced NSCLC (34,35) as it can significantly enhance patient survival rates. Neoadjuvant combination therapy may be offered to patients in the intermediate stage, thereby substantially extending progression-free survival. Patients responsive to immunotherapy often exhibit characteristics such as elevated programmed death ligand-1 (PD-L1) expression (36), high TMB (37,38), KRAS mutations, and co-mutations of KRAS and TP53. However, these tests alone may not fully meet patient needs, owing to discrepancies in their accuracy and sensitivity. Furthermore, the clinical utility of STK11, KEAP1, and EGFR gene alterations in guiding ICB therapy for NSCLC remains controversial (39). This finding is of considerable significance given the well-documented limitations associated with PD-L1, TMB, and the genes STK11 and KEAP1. Although PD-L1 expression is commonly employed to predict responses to PD-1/PD-L1 inhibitors, it is not without its limitations. In contrast, ACSM5 provides a novel perspective. Our research indicates that ACSM5 modulates the immune response through lipid metabolism, a mechanism distinct from the PD-L1-regulated immune checkpoint pathway. TMB does not directly account for immune cell metabolism, whereas ACSM5 is concerned with the energy supply. Even in tumors with high TMB, immune responses may be compromised if immune cells are deprived of energy due to ACSM5-related lipid metabolism issues. Thus, ACSM5 could enhance the predictive accuracy of immunotherapy responses in conjunction with TMB. While STK11 and KEAP1 are implicated in metabolism and oxidative stress signaling, ACSM5 exerts a more direct influence on lipid metabolism, which is a crucial energy source for immune cells. Our study demonstrates that low ACSM5 expression results in T-cell exhaustion, a significant contributor to immunotherapy resistance. Future studies will be conducted to further elucidate the mechanisms by which ACSM5 influences the outcomes of immunotherapeutic treatments in NSCLC. Consequently, identification of novel and effective targets for immunotherapy has become increasingly crucial. This research began with individuals undergoing immunotherapy, aiming to identify genes associated with immunotherapy response, construct a prognostic signature, and subsequently validate the model. Differential analysis was conducted on the immune microenvironment, immune checkpoints, immune function, and immunotherapy responses between the high-risk and low-risk groups of the model.

A limitation of this study is the insufficient number of immunotherapy samples in the database, which resulted in a restricted pool of samples available for analysis. WGCNA exhibited a relatively low correlation owing to this constraint. Fortunately, the p-values were exceedingly low, which could be attributed to the statistical significance of the results. To address the issue of inadequate data, we strategically integrated these genes into TCGA and GEO datasets, leveraging a larger sample size to conduct analysis of the selected genes, ensuring the robustness and feasibility of future investigations. This analytical method has limitations due to heterogeneity between LUSC and LUAD. Future studies with larger, diverse groups will verify the biomarkers’ relevance for both LUAD and LUSC.

ACSM5 is a protein-coding gene (40). The role of ACSM5 in tumor progression is still in the initial stage of research, and current research indicates that it is a gene linked to lipid metabolism. Both thyroid cancer and LUAD have been reported to express this gene, and it is also expressed at low levels in thyroid cancer. Research has indicated that ACSM5 affects the FABP4 signaling pathway (41), which in turn affects lipid metabolism (42,43). In this study, we found that ACSM5 was expressed at low levels in tumors. This may be attributed to ACSM5’s attenuation of lipid metabolism function, which fails to provide sufficient energy for immune cells, leading to T cell exhaustion and poor immunotherapy outcomes (44,45). ACSM5 is a prognostic factor in NSCLC and is strongly associated with the success of immunotherapy according to this study, which utilized advanced bioinformatics analysis. This finding is significant because there are currently very few studies of ACSM5. Moving forward, we will conduct further studies to investigate the mechanisms by which ACSM5 affects the outcomes of immunotherapeutic treatment in NSCLC.

Conclusions

In summary, we identified a prognostic marker for NSCLC and established an association between this gene and the effectiveness of immunotherapy using modern and reliable bioinformatics research. Experimental validation was performed to determine the expression of this gene in NSCLC. This study has the potential to offer substantial recommendations for clinical immunotherapy.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1620/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1620/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1620/prf

Funding: This study was supported by the Tianjin Key Medical Discipline (Specialty) Construction Project (No. TJYXZDXK-010A); Tianjin Medical University Cancer Institute and Hospital’s Foundation for Talent Recruitment and Doctor (No. B2414); and Tianjin Medical University Cancer Institute and Hospital’s Special Foundation for Pharmaceutical, Laboratory, and Imaging Sciences (No. Y2307).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1620/coif). All authors report the funding from the Tianjin Key Medical Discipline (Specialty) Construction Project (No. TJYXZDXK-010A); Tianjin Medical University Cancer Institute and Hospital’s Foundation for Talent Recruitment and Doctor (No. B2414); and Tianjin Medical University Cancer Institute and Hospital’s Special Foundation for Pharmaceutical, Laboratory, and Imaging Sciences (No. Y2307). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethical Board of the Tianjin Medical University Cancer Institute and Hospital (No. bc2023152), and informed consent was obtained from all participating patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Yiyan S, Yang S, Li D, et al. Vitamin D Affects the Warburg Effect and Stemness Maintenance of Non- Small-Cell Lung Cancer Cells by Regulating the PI3K/AKT/mTOR Signaling Pathway. Curr Cancer Drug Targets 2022;22:86-95. [Crossref] [PubMed]
2. Wood C, Lyniv L, Isaacs JM, et al. Perioperative pembrolizumab in early-stage non-small cell lung cancer (NSCLC): safety, efficacy, and exploratory biomarker analysis. J Immunother Cancer 2025;13:e010395. [Crossref] [PubMed]
3. Wang Q, Zeng A, Zhu M, et al. Dual inhibition of EGFR‑VEGF: An effective approach to the treatment of advanced non‑small cell lung cancer with EGFR mutation Int J Oncol 2023;62:26. (Review). [Crossref] [PubMed]
4. Hendriks LEL, Remon J, Faivre-Finn C, et al. Non-small-cell lung cancer. Nat Rev Dis Primers 2024;10:71. [Crossref] [PubMed]
5. Kong C, Chen X. Combined Photodynamic and Photothermal Therapy and Immunotherapy for Cancer Treatment: A Review. Int J Nanomedicine 2022;17:6427-46. [Crossref] [PubMed]
6. de Jong C, Chargi N, Herder GJM, et al. The association between skeletal muscle measures and chemotherapy-induced toxicity in non-small cell lung cancer patients. J Cachexia Sarcopenia Muscle 2022;13:1554-64. [Crossref] [PubMed]
7. Dejonckheere CS, Nour Y, Sahlmann J, et al. Impact of estimated dose of radiation to immune cells (EDRIC) in locally advanced Non-Small-Cell lung Cancer: A secondary analysis of the multicenter randomized PET-Plan trial. Radiother Oncol 2025;208:110907. [Crossref] [PubMed]
8. Dong C, Chen Y, Li H, et al. The antipsychotic agent flupentixol is a new PI3K inhibitor and potential anticancer drug for lung cancer. Int J Biol Sci 2019;15:1523-32. [Crossref] [PubMed]
9. Fu S, Li G, Zang W, et al. Pure drug nano-assemblies: A facile carrier-free nanoplatform for efficient cancer therapy. Acta Pharm Sin B 2022;12:92-106. [Crossref] [PubMed]
10. Li J, Burgess DJ. Nanomedicine-based drug delivery towards tumor biological and immunological microenvironment. Acta Pharm Sin B 2020;10:2110-24. [Crossref] [PubMed]
11. Leko V, Cafri G, Yossef R, et al. Identification of neoantigen-reactive T lymphocytes in the peripheral blood of a patient with glioblastoma. J Immunother Cancer 2021;9:e002882. [Crossref] [PubMed]
12. Fortuny M, Sanduzzi-Zamparelli M, Reig M. Systemic therapies in hepatocellular carcinoma: A revolution? United European Gastroenterol J 2024;12:252-60. [Crossref] [PubMed]
13. Fan Y, Liu Y, Wang L, et al. bITH, a blood-based metric of intratumor heterogeneity, is associated with clinical response to immune checkpoint blockade in non-small cell lung cancer. EBioMedicine 2023;91:104564. [Crossref] [PubMed]
14. Almeida JS, Sousa LM, Couceiro P, et al. AI-assisted peripheral immune profiling reveals unconventional lymphocyte signatures associated with prognosis in soft tissue sarcoma patients. Front Immunol 2025;16:1677408. [Crossref] [PubMed]
15. Shu J, Jiang J, Zhao G. Identification of novel gene signature for lung adenocarcinoma by machine learning to predict immunotherapy and prognosis. Front Immunol 2023;14:1177847. [Crossref] [PubMed]
16. Shibaki R, Fujimoto D, Nozawa T, et al. Machine learning analysis of pathological images to predict 1-year progression-free survival of immunotherapy in patients with small-cell lung cancer. J Immunother Cancer 2024;12:e007987. [Crossref] [PubMed]
17. Zhang Z, Chen L, Chen H, et al. Pan-cancer landscape of T-cell exhaustion heterogeneity within the tumor microenvironment revealed a progressive roadmap of hierarchical dysfunction associated with prognosis and therapeutic efficacy. EBioMedicine 2022;83:104207. [Crossref] [PubMed]
18. Chen X, Ma J. Weighted gene co-expression network analysis (WGCNA) to explore genes responsive to Streptococcus oralis biofilm and immune infiltration analysis in human gingival fibroblasts cells. Bioengineered 2021;12:1054-65. [Crossref] [PubMed]
19. Xu L, Zhao F, Ren H, et al. Co-expression analysis of fetal weight-related genes in ovine skeletal muscle during mid and late fetal development stages. Int J Biol Sci 2014;10:1039-50. [Crossref] [PubMed]
20. Feng S, Xu Y, Dai Z, et al. Integrative Analysis From Multicenter Studies Identifies a WGCNA-Derived Cancer-Associated Fibroblast Signature for Ovarian Cancer. Front Immunol 2022;13:951582. [Crossref] [PubMed]
21. Lv JH, Hou AJ, Zhang SH, et al. WGCNA combined with machine learning to find potential biomarkers of liver cancer. Medicine (Baltimore) 2023;102:e36536. [Crossref] [PubMed]
22. Rezaei Z, Ranjbaran J, Safarpour H, et al. Identification of early diagnostic biomarkers via WGCNA in gastric cancer. Biomed Pharmacother 2022;145:112477. [Crossref] [PubMed]
23. Chen Y, Zhou C, Zhang X, et al. Construction of a novel radioresistance-related signature for prediction of prognosis, immune microenvironment and anti-tumour drug sensitivity in non-small cell lung cancer. Ann Med 2025;57:2447930. [Crossref] [PubMed]
24. Zimmermann SC, Tonga KO, Thamrin C. Dismantling airway disease with the use of new pulmonary function indices. Eur Respir Rev 2019;28:180122. [Crossref] [PubMed]
25. Qin H, Abulaiti A, Maimaiti A, et al. Integrated machine learning survival framework develops a prognostic model based on inter-crosstalk definition of mitochondrial function and cell death patterns in a large multicenter cohort for lower-grade glioma. J Transl Med 2023;21:588. [Crossref] [PubMed]
26. Liang P, Yang W, Chen X, et al. Machine Learning of Single-Cell Transcriptome Highly Identifies mRNA Signature by Comparing F-Score Selection with DGE Analysis. Mol Ther Nucleic Acids 2020;20:155-63. [Crossref] [PubMed]
27. Daidone M, Ferrantelli S, Tuttolomondo A. Machine learning applications in stroke medicine: advancements, challenges, and future prospectives. Neural Regen Res 2024;19:769-73. [Crossref] [PubMed]
28. Zhang G, Tang Q, Feng P, et al. IPs-GRUAtt: An attention-based bidirectional gated recurrent unit network for predicting phosphorylation sites of SARS-CoV-2 infection. Mol Ther Nucleic Acids 2023;32:28-35. [Crossref] [PubMed]
29. Stagno F, Russo S, Murdaca G, et al. Utilization of Machine Learning in the Prediction, Diagnosis, Prognosis, and Management of Chronic Myeloid Leukemia. Int J Mol Sci 2025;26:2535. [Crossref] [PubMed]
30. Zhang L, Cui Y, Mei J, et al. Exploring cellular diversity in lung adenocarcinoma epithelium: Advancing prognostic methods and immunotherapeutic strategies. Cell Prolif 2024;57:e13703. [Crossref] [PubMed]
31. Zhang L, Cui Y, Zhou G, et al. Leveraging mitochondrial-programmed cell death dynamics to enhance prognostic accuracy and immunotherapy efficacy in lung adenocarcinoma. J Immunother Cancer 2024;12:e010008. [Crossref] [PubMed]
32. Pal S, Bi Y, Macyszyn L, et al. Isoform-level gene signature improves prognostic stratification and accurately classifies glioblastoma subtypes. Nucleic Acids Res 2014;42:e64. [Crossref] [PubMed]
33. Zhou Z, Zhang J, Xu C, et al. An integrated model of N6-methyladenosine regulators to predict tumor aggressiveness and immune evasion in pancreatic cancer. EBioMedicine 2021;65:103271. [Crossref] [PubMed]
34. Xu Y, Xu J, Qiao R, et al. Loss of BLK expression as a potential predictor of poor prognosis and immune checkpoint blockade response in NSCLC and contribute to tumor progression. Transl Oncol 2023;33:101671. [Crossref] [PubMed]
35. Kim H, Park S, Han KY, et al. Clonal expansion of resident memory T cells in peripheral blood of patients with non-small cell lung cancer during immune checkpoint inhibitor treatment. J Immunother Cancer 2023;11:e005509. [Crossref] [PubMed]
36. Ribas A, Hu-Lieskovan S. What does PD-L1 positive or negative mean?. J Exp Med 2016;213:2835-40. [Crossref] [PubMed]
37. Jung J, Heo YJ, Park S. High tumor mutational burden predicts favorable response to anti-PD-(L)1 therapy in patients with solid tumor: a real-world pan-tumor analysis. J Immunother Cancer 2023;11:e006454. [Crossref] [PubMed]
38. Xiong Y, Wang Z, Zhou Q, et al. Identification and validation of dichotomous immune subtypes based on intratumoral immune cells infiltration in clear cell renal cell carcinoma patients. J Immunother Cancer 2020;8:e000447. [Crossref] [PubMed]
39. van de Haar J, Mankor JM, Hummelink K, et al. Combining Genomic Biomarkers to Guide Immunotherapy in Non-Small Cell Lung Cancer. Clin Cancer Res 2024;30:1307-18. [Crossref] [PubMed]
40. Ma C, Luo H, Cao J, et al. Identification of a Novel Tumor Microenvironment-Associated Eight-Gene Signature for Prognosis Prediction in Lung Adenocarcinoma. Front Mol Biosci 2020;7:571641. [Crossref] [PubMed]
41. Cao Y, Li J, Qiu S, et al. ACSM5 inhibits ligamentum flavum hypertrophy by regulating lipid accumulation mediated by FABP4/PPAR signaling pathway. Biol Direct 2023;18:75. [Crossref] [PubMed]
42. Yang L, Pham K, Xi Y, et al. Acyl-CoA Synthetase Medium-Chain Family Member 5-Mediated Fatty Acid Metabolism Dysregulation Promotes the Progression of Hepatocellular Carcinoma. Am J Pathol 2024;194:1951-66. [Crossref] [PubMed]
43. Rinaldi A, Lazareth H, Poindessous V, et al. Impaired fatty acid metabolism perpetuates lipotoxicity along the transition to chronic kidney injury. JCI Insight 2022;7:e161783. [Crossref] [PubMed]
44. Zhang Y, Kurupati R, Liu L, et al. Enhancing CD8(+) T Cell Fatty Acid Catabolism within a Metabolically Challenging Tumor Microenvironment Increases the Efficacy of Melanoma Immunotherapy. Cancer Cell 2017;32:377-391.e9. [Crossref] [PubMed]
45. Mansouri A, Gattolliat CH, Asselah T. Mitochondrial Dysfunction and Signaling in Chronic Liver Diseases. Gastroenterology 2018;155:629-47. [Crossref] [PubMed]