Predictive value of immunomodulatory-related genes for the prognosis of lung adenocarcinoma based on bioinformatics analysis Authors: Luyu Yang, Zhimin Cao, Xing Yuanfang, Teng Ma, Huan Ye, Dongchang Wang Published: 2025 ABSTRACT AbstractLung adenocarcinoma (LUAD), characterized by its heterogeneity and high invasiveness, remains a challenging disease despite significant advancements in immunotherapy. Our study aimed to elucidate the expression differences of immunomodulatory-related genes in LUAD tissues compared to normal tissues and to validate the predictive role of these genes in the diagnosis, immunotherapy, and prognostic assessment of LUAD. Gene expression profiles and clinical data were extracted from TCGA and GEO datasets. GO analysis, KEGG pathway analysis, and GSEA were employed to identify biological process disparities. The single-sample Gene Set Enrichment Analysis (ssGSEA) was used to assess immune cell infiltration in distinct LUAD disease subtypes and derive an immunomodulatory score (Is) for patients with LUAD. Univariate and multivariate Cox regression analyses were performed to ascertain the prognostic significance of immunomodulatory-related genes (IRGs) in LUAD. Calibration curves and decision curve analysis (DCA) was used to verify the predictive efficacy of these IRs for the prognosis of patients with LUAD. This study identified, in the TCGA dataset, 18 IRGs that were differentially expressed between the LUAD and normal groups. Differential expression analysis IRGs using a threshold of |logFC|> 1 with an adjusted P-value < 0.05. GO and KEGG enrichment analysis, as well as GSEA results, revealed several significantly upregulated signaling pathways, including oxidative-stress-induced cell senescence and cell response to hypoxia. Interestingly, a strong correlation was observed between IRGs and activated dendritic cells in the four LUAD disease subtypes. Cox regression analyses demonstrated that the clinical prognosis of patients with LUAD was significantly associated with several key prognostic genes and clinical variables, including T, N, and M stages. The calibration curve and DCA verify this result. In summary, IRGs are increasingly being recognized for their pivotal roles in LUAD progression. The LUAD disease subtypes and Is derived from the IRGs offer valuable tools for predicting immunotherapy prognosis and stratifying the LUAD diagnosis.Graphical Abstract Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-025-04000-1.Keywords: Immune-related differentially expressed genes, Lung Adenocarcinoma, Bioinformatics, Diagnosis, Prognosis Introduction Despite significant advancements in cancer screening and treatment modalities, lung adenocarcinoma (LUAD) cancer remains one of the most serious causes of death worldwide, whereas Adenocarcinoma is the most common and aggressive histological subtype of all non-small cell lung cancers [1–6]. In recent years, in pace with the progress in molecular biology, immune checkpoint inhibitor (ICI) therapy has become a new first-line antitumor treatment [7]. ICI reactivates the cytotoxic properties of T cells by targeting CTLA-4 and PD-1, thereby effectively completing the tumor antigen-specific immune response and killing the tumor, resulting in long-lasting benefits for patients [8]. However, ICI monotherapy could only benefit a few patients, with objective response rates of only 15–25%, which might be attributed to tumor heterogeneity and tumor drug resistance [9]. In addition, studies have shown that using ICIs might disrupt immune homeostasis, reduce T-cell tolerance or increase the content of preexisting autoantibodies and inflammatory factors, leading to adverse events [10]. Therefore, although immunotherapy has excellent prospects for prolonging the survival of patients with LUAD, the 5-year survival rate of these patients is still at 19% [11]. Hence, there is a need to explore alternative immune checkpoints for anticancer therapy. Further elucidation of the molecular mechanisms of LUAD progression will help identify safer and more effective cancer biomarkers or potential prognostic factors. In recent years, with the advances in bioinformatics and genomics, the roles of immunomodulatory-related genes (IRGs) in LUAD have gradually received attention [12, 13]. IRGs are a class of genes that encode or regulate the function of the immune system. They play an important role in the development and progression of cancer. Studies have shown, for example, that the metastasis and invasion of cervical cancer are mainly attributed to the tumor microenvironment (TME), mainly composed of infiltrating immune cells and stromal cells, and genes related to immune regulation play a significant role in immune activities [14–16]. Moreover, recent research has delved into IRGs through bioinformatic analyses, revealing that IRGs have potential as prognostic markers and therapeutic targets in several cancers, including cervical, colon, bladder, and pancreatic cancers [17]. However, the precise role of IRGs in determining the prognosis of LUAD remains elusive. Thus, our study sought to leverage bioinformatics analysis to identify IRGs specific to LUAD, with the ultimate goal of developing novel strategies for the precise treatment and prognosis evaluation of this disease. This study utilized a TCGA-LUAD dataset as the test set, complemented by GSE30219, GSE10072, GSE3141 as the validation sets for our analysis. Initially, we conducted a differential analysis of the gene expression profiles within the samples from the TCGA-LUAD dataset (classified into LUAD and normal groups). Subsequently, we intersected the resulting differentially expressed genes (DEGs) with a previously specified group of IRGs to identify the immediate-early genes (IEGs). Based on these IEGs, we identified four subtypes of LUAD and demonstrated their predictive value for predicting the success rate of immunotherapies targeting PD-1 and CTLA-4. Furthermore, we constructed an immunomodulatory score (Is) based on IEGs and validated its ability to differentiate between the LUAD and normal groups. This study also constructed a prognostic model with six key prognostic genes in IEGs and investigated how these key prognostic genes, along with clinical variables such as T, N, and M stages, could act as independent prognostic elements for patients with LUAD. Finally, column charts and calibration curves were plotted to assess the predictive ability of the prognostic model. Our results showed that the model had a good predictive effect on the 5-year survival rate of patients with LUAD and could be applied to predict their clinical prognosis. Therefore, this study provides new biomarkers for guiding the diagnosis, treatment, and prognostic evaluation of LUAD, which might bring new hope for its treatment. Materials and methods Data acquisition and ethical compliance The TCGA and GEO datasets used in this study are publicly available repositories from which all personal identifiers have been removed (de-identified). According to the official data-use policies of TCGA (https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga/history) and GEO (https://www.ncbi.nlm.nih.gov/geo/info/disclaimer.html), all data have undergone relevant ethical review and are openly provided for use by researchers worldwide. Secondary analyses of such de-identified, population-level genomic data generally do not require separate ethical approval or an ethics-waiver number. This study strictly complies with the usage guidelines of these databases to ensure lawful and appropriate use of the data. Data download The TCGA-LUAD dataset was obtained from the Cancer Genome Atlas via TCGABiolinks and analyzed as a test set. After excluding data samples that did not include any clinical information, 598 samples containing clinical information were collected [18]. The dataset comprised 59 normal paracarcinoma samples (grouping: normal group) and 539 LUAD samples. The count sequencing data of LUAD was also standardized to the FPKM format, while the clinical data corresponding to samples was acquired from the UCSC Xena database [19]. Gene somatic mutation data (masked somatic mutations) of the TCGA-LUAD dataset were downloaded from TCGA. Additionally, the tumor mutation burden (TMB) data and microsatellite instability (MSI) data were downloaded from the cBioPortal [20]. We also downloaded the expression profile datasets GSE30219, GSE10072, and GSE3141 of patients with LUAD from the GEO database using the R package GEOquery [21–25]. All three GSE30219, GSE10072, and GSE3141 datasets belong to Homo sapiens. The GSE30219 dataset was obtained from the GPL570 [HG-U133_plus_2] Affymetrix Human Genome U133 Plus 2.0 array, which includes microarray gene expression profiles from 307 samples. Among these samples, 83 were from LUAD tumor tissues (grouping: LUAD group) and 14 were from non-LUAD tumor tissues (grouping: normal group); they were all included in this analysis. The GSE10072 dataset was obtained from the GPL96 [HG-U133A] Affymetrix Human Genome U133A array, which comprises 180 microarray gene expression profiles, including 58 LUAD and 49 non-neoplasm tissue samples. From this array, a total of 107 samples were included in the analysis. Similarly, the GSE3141dataset was obtained from the GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 array and comprises 111 microarray genetic expression profiles of frozen primary LUAD tumor tissues. Samples with a survival time of less than one month were excluded, resulting in 110 samples being included in this analysis. The GSE30219, GSE10072, and GSE3141 datasets were later used as the validation sets (Table 1). Table 1Lung adenocarcinoma dataset information listTCGA-LUADGSE30219GSE10072GSE3141Platform/GPL570GPL96GPL570SpeciesHomo sapiensHomo sapiensHomo sapiensHomo sapiensTissueLungsLungsLungsLungsSamples in normal group5914490Samples in LUAD group5398358110Reference/[6][7][8]TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma Lung adenocarcinoma dataset information list TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma Additionally, we collected IRGs through the GeneCards database [26]. Using the search term “Immunomodulatory,” we acquired a cumulative count of 171 IRGs. Through a comprehensive literature review, we identified 2 other IRGs, NCOA4 and TNFRSF4 [27, 28]. We obtained 172 IRGs by merging the IRGs from the two sources. These 172 IRGs were then intersected with the genes present in the TCGA-LUAD, GSE30219, GSE10072, and GSE3141 datasets. This intersection provided a final count of 158 IRGs for subsequent analysis. The names of these genes are listed in Table S1. Differentially expressed genes (DEGs) We first standardized the TCGA-LUAD, GSE30219, GSE10072, and GSE3141 tumor datasets using the limma package [29]. We also used this limma package to analyze the expression profile data in TCGA-LUAD and identify DEGs between the two groups (LUAD and normal). DEGs were classified based on the following criteria: if log-fold change (logFC) > 1 and P adjust < 0.05, the DEGs were considered upregulated; but, if |logFC|< 1 and P adjust < 0.05, DEGs were considered as downregulated. Subsequently, we intersected all |logFC|> 1 and P adjust < 0.05 DEGs from the TCGA-LUAD dataset with the 158 IRGs previously selected and plotted a Venn diagram to obtain the IEGs. A volcano plot was created using ggplot2, and a heatmap was generated using pheatmap both in R [30, 31]. LUAD disease subtype identification through single nucleotide polymorphism (SNP) and copy number variation (CNV) gene analyses We used VarScan to preprocess the somatic mutation data of patients with LUAD in the TCGA-LUAD dataset and the maftools package to visualize somatic mutations in the IEGs [32]. CNV is a phenomenon involving genomic alterations that can lead to one or more gene copy number abnormalities. The “Masked Copy number Segment” from the TCGA-LUAD dataset was obtained using TCGABiolinks in R, and a total of 619 CNV cases were identified in patients with LUAD. CNV was visually analyzed using ggplot2. Functional and pathway enrichment analyses Gene Ontology (GO) analysis is a common methodology for conducting extensive functional enrichment studies [33]; it comprises several categories, including biological process (BP), molecular function (MF), and cellular component (CC) categories. The Kyoto Encyclopedia of Genes and Genomes (KEGG) is a widely used database for storing information about genomes, biological pathways, diseases, and drugs [34]. We used the R package clusterProfiler to conduct GO and KEGG annotation analyses of the identified IEGs between the LUAD and normal groups in TCGA-LUAD [35]. The following criteria were used for screening: P adjust < 0.05 and false discovery rate (FDR) (q value) < 0.25, which were considered statistically significant. The P value was adjusted using the Benjamini-Hochberg (BH) method. Gene enrichment analysis using GSEA GSEA is used to assess the distribution trend of a predefined set of genes in correlation with specific phenotypes and, thus, to determine their contribution to such phenotypes [36]. In this study, the genes in the TCGA-LUAD dataset were divided into high- and low-phenotypic relevance groups based on their phenotypic relevance rankings. ClusterProfiler was then used to perform an enrichment analysis of the genes shared by the LUAD and normal groups. The GSEA parameters were defined as follows: seed for permutation: 2022; 1000 permutations; and each gene set contained at least 10 genes and a maximum of 500. We applied the BH method for P value correction. The file obtained was named c2.cp.all.v2022.1. The Hs.symbols.gmt gene set from the MSigDB database was used as a reference gene, and the screening criteria for significant enrichment were P adjust < 0.05 and FDR (q value) < 0.05. LUAD disease subtypes classification Consensus Clustering is a resampling-based algorithm used to identify samples and their subgroups and verify the validity of the discovered clusters [37]. Consistency clustering is a multiple iteration method on subsamples of a dataset that provides an index of clustering stability and parameter decision-making by inducing variation in sample taking through the use of such subsamples. The ConsensusClusterPlus package in R was used to identify the varying LUAD disease subtypes stemming from IEGs. During this process, we set the number of clusters between 2 and 8 and a 50-times repetition to extract 80% of the total sample; other parameters included clusterAlg = “km” and distance = “euclidean [38].” The differences in the expression of IEGs among the different LUAD disease subtypes in the TCGA-LUAD dataset were calculated using the Mann–Whitney U test; P values < 0.05 were considered to indicate statistically significant differences. Analysis of immune infiltration and immunotherapy based on disease subtype grouping To further quantify the proportions of different immune cells in LUAD samples, we used the ssGSEA algorithm. This approach offers high sensitivity and specificity for distinguishing between various human immune cell phenotypes within the tumor immune microenvironment. Using this algorithm, we obtained 28 curated gene sets based on the seminal work by Bindea et al. to characterize diverse tumor-infiltrating immune cell populations. These gene signatures encompass multiple human immune cell subtypes, including CD8+ T cells, dendritic cells (DCs), macrophages, and regulatory T cells (Tregs). [39]. The enrichment scores obtained using ssGSEA in the gene set variation analysis (GSVA) package in R were used to represent the infiltration level of each immune cell type in each sample. Differences in the infiltration abundance of immune cells among cluster1, cluster2, cluster3, and cluster4 of the LUAD group of the TCGA-LUAD dataset were represented by box plots. Besides, the relevance of different immune cells within LUAD disease subgroups was calculated using the Spearman algorithm and visualized through ggplot2. We then combined the TCGA-LUAD dataset gene expression matrix to calculate the correlation between immune cells and IEGs among different LUAD disease subgroups and demonstrated the correlation by drawing a heatmap using pheatmap. Immunogenicity refers to the ability of an antigen or its epitope to act on the antigen-recognizing receptors of T and B cells, thereby inducing a humoral or cell-mediated immune response in the body. Immunogenicity is governed by multiple genes, including those related to effector cells, MHC molecules, immunomodulatory factors, and immunosuppressive cells. Machine learning can be used to estimate and quantify immunogenicity. The TCIA database provides the immunophenoscores (IPS) of 20 different cancers, which might predict CTLA-4 and PD-1 reactivity [40]. Therefore, the IPS data of LUAD were obtained from the TCIA database and analyzed using ggplot2. Determination of the immunomodulatory score (Is) The ssGSEA algorithm quantifies the relative abundance of the genes present in dataset samples [41]. This algorithm was used to calculate the Is of each sample in the TCGA-LUAD and GSE3141 datasets based on the expression matrix of the IEGs. According to the median grouping method, we obtained two groups, a high-Is group, and a low-Is group, from the TCGA-LUAD and GSE3141 datasets. Then, the Is differences between the LUAD and normal groups from both the TCGA-LUAD and GSE3141 datasets, as well as the differences between the High and Low-Is groups from the TCGA-LUAD dataset, were calculated using the Mann–Whitney U test (Wilcoxon rank sum test). Statistical significance was set at P < 0.05. GSVA enrichment differences between the high and low-Is groups GSVA is a non-parametric and unsupervised analysis method that is primarily used to evaluate gene set enrichment differences between the nuclear transcriptomes of different samples. GSVA transforms gene expression matrices into gene set expression matrices to determine whether different pathways are enriched among different samples [42]. We used "h.all.v7.4. symbols.gmt" (from MSigDB) as the reference gene set, and GSVA was performed on the expression matrix of the TCGA-LUAD and GSE3141 datasets [43]. The discrepancies in the function of the enriched pathways obtained from the GSVA enrichment analysis between the high and low-Is groups of LUAD were analyzed, and P values < 0.05 were considered statistically significant. Constructing a prognostic risk model based on IEGs To construct an IEG-based prognostic model, we used glmnet with the IEGs from the TCGA-LUAD dataset as load data; glmnet was run with the random seed number parameter set to “2022” and including a LASSO regression with tenfold cross-validation-which was employed for regularization and dimensionality reduction analysis-to further screen for the prognostic IEGs [44, 45]. The LASSO regression is commonly used to construct prognostic models; it is based on linear regressions and reduces overfitting by adding a penalty term (lambda × absolute value of the slope) while improving the model's generalization ability. Lambda‐selection criterion: Deviance was used as the metric to determine the optimal λ.Cross-validation method: tenfold cross-validation was employed, and the entire procedure was repeated 10 times to ensure robustness.The random seed was set to 2022 to guarantee reproducibility.We used the median value of the IEGs prognostic model risk score to classify the patients in the TCGA-LUAD dataset into high- and low-risk groups. We then visualized the LASSO regression results using a risk factor map. Further, we illustrated the grouping, survival outcome, and molecular expression of the prognostic IEGs in each sample according to the risk score in the LASSO regression prognostic model. Clinical evaluation of prognostic IEGs To investigate the clinical prognostic value of the identified prognostic IEGs in LUAD, a univariate Cox regression analysis was performed on the expression of these prognostic IEGs and the clinical variables in the TCGA-LUAD dataset, and those factors with P < 0.1 were used to construct a multivariate Cox regression model. Subsequently, we used the RMS function of the R package to construct a nomogram of the results; nomograms are graphs that use a cluster of non-intersecting line segments to represent the functional relationships among multiple independent variables in a Cartesian plane coordinate system [46]. Based on multifactor regression analyses, we established a group of scales to score and characterize each variable. Finally, the total score was calculated to forecast the odds of an event occurring. A calibration curve was then used to evaluate the accuracy and resolution of the nomogram [47]. It also evaluated the predictive performance of the model with actual results by plotting the fit between the actual and predicted probabilities under different circumstances. Calibration curves are mainly used to analyze the degree of fit between a model established by Cox regression and its practicality. Decision curve analysis (DCA) is a simple method for evaluating clinical predictive models, diagnostic tests, and molecular markers [48]. Finally, we used DCA graph drawing by ggDCA in the R package to assess the exactitude and resolution of the Cox regression model. Statistical analyses R software (version 4.1.2) was used to analyze all data in this study. For the comparison of two consecutive variables, the independent Student’s t-test was used to estimate the statistical significance of normally distributed variables, and the difference between non-normally distributed variables was analyzed using the Mann–Whitney U test. Three or more groups were compared using the Kruskal–Wallis test. Either the chi-square or Fisher’s exact tests were used to compare and analyze the statistical significance between two sets of categorical variables. The results were calculated using Spearman’s correlation analysis and the correlation coefficients between different molecules. The survival package in R was used for survival analysis. The Kaplan–Meier survival curve was used to display survival discrepancy, and the log-rank test was used to evaluate the significance of survival time differences between the two groups. Unless specifically indicated, all statistical P values were bilateral, and P < 0.05 was considered statistically significant. Software version information All analyses were performed in R version 4.3.2 (2023–10-31). Key R package versions are listed below: Data acquisition and preprocessing TCGAbiolinks (v2.36.0). TCGAbiolinks (v2.36.0). GEOquery (v2.76.0). GEOquery (v2.76.0). Statistical analysis limma (v3.64.0). limma (v3.64.0). glmnet (v4.1.8). glmnet (v4.1.8). survival (v3.8.3). survival (v3.8.3). rms (v8.0.0). rms (v8.0.0). Functional enrichment clusterProfiler (v4.16.0). clusterProfiler (v4.16.0). GSVA (v2.2.0). GSVA (v2.2.0). Clustering and visualization ConsensusClusterPlus (v1.72.0). ConsensusClusterPlus (v1.72.0). ggplot2 (v3.5.2). ggplot2 (v3.5.2). pheatmap (v1.0.12). pheatmap (v1.0.12). RCircos (v1.2.2). RCircos (v1.2.2). ggDCA (v1.1). ggDCA (v1.1). Genomic analysis maftools (v2.24.0). Results Technology roadmap The analysis process of this study is shown in Fig. 1. Technology Roadmap TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. Is, Immunomodulatory score; IRGs, Immunomodulatory-related genes; IRDEGs, Immunomodulatory-related differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; GSEA, Gene Set Enrichment Analysis; GSVA, Gene Set Variation Analysis; SNP, Single nucleotide polymorphism; CNV, Copy number variation; LASSO, Least absolute shrinkage and selection operator Differences in the expression of IEGs in the LUAD dataset To obtain the IEGs, we divided the samples in the TCGA-LUAD dataset into LUAD and normal groups and conducted a differential analysis of the data. A total of 19,938 DEGs were identified, and only 1674 of these DEGs fulfilled the preestablished classification criteria. Under such criteria, 706 DEGs were upregulated in the LUAD group (showing a lower expression in the normal group, logFC > 1, P adjust < 0.05), and 968 DEGs were downregulated (showing a higher expression in the normal group, logFC < -1, P adjust < 0.05). A volcano map was constructed using these results (Fig. 2A). To identify the IEGs, the 1674 DEGs from the TCGA-LUAD dataset that fulfilled the selection criteria were intersected with the 158 previously selected IRGs. A total of 18 IEGs (BMP6, CAT, CD79A, CD83, CSF3, CXCL12, CXCL2, CXCR2, HLA-E, HSPD1, HSPE1, IL6, PPARG, S1PR1, S1PR4, SELE, TIMP1, TLR4) were obtained, and a Venn diagram was drawn using these data (Fig. 2B). Expression difference of IRDEGs in LUAD dataset. A Volcano map display of differential expression analysis between LUAD group and Normal group in TCGA-LUAD dataset. B Venn diagram display of DEGs and 158 IRGs of the differential expression analysis results between the LUAD group and the Normal group in the TCGA-LUAD dataset. C Volcano map display of IRDEGs between the LUAD group and the normal group in the TCGA-LUAD dataset. D-F. PCA plots of IRDEGs in TCGA-LUAD dataset (D), GSE30219 dataset (E), and GSE10072 dataset (F). G-I. IRDEGs in the TCGA-LUAD dataset (G), GSE30219 dataset (H), and GSE10072 (I) dataset expression difference analysis results are shown in group comparison charts. The symbol ns is equivalent to P ≥ 0.05, no statistical significance; the symbol * is equivalent to P < 0.05; the symbol ** is equivalent to P < 0.01; the symbol *** is equivalent to P < 0.001. IRGs, Immunomodulatory-related genes. DEGs, differentially expressed genes. IRDEGs, Immunomodulatory-related differentially expressed genes. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. PCA, Principal Component Analysis We then analyzed the differential expression of the 18 IEGs in the LUAD and normal groups from TCGA-LUAD, and draw a heatmap to visualize the results. In addition, we conducted a principal component analysis on the expression matrices of TCGA-LUAD, GSE30219 dataset and GSE10072 to determine the distribution of LUAD samples in the three datasets (Fig. 2D–F). The results showed dramatic diversity between samples from different groups in the three LUAD datasets. We then extracted the microarray chip data of the 18 IEGs from the three datasets. The Mann–Whitney U test was used to analyze the expression differences of the 18 IEGs in LUAD tumor (grouping: LUAD) and normal (grouping: normal) tissues in the three LUAD datasets, and the results of a relevance analysis of expression discrepancy by groups were visualized using comparison plots (Fig. 2G–I). The results showed that, in the TCGA-LUAD dataset, the expression of all 18 IEGs was statistically different between the two groups (P < 0.05); among them, genes CD79A, HSPD1, HSPE1 and TIMP1 in the LUAD group were significantly upregulated compared to those in the normal group (P < 0.05). In addition, genes BMP6, CAT, CD83, CSF3, CXCL12, CXCL2, CXCR2, HLA-E, IL6, PPARG, S1PR1, S1PR4, SELE, TIMP1, TLR4 in group LUAD were significantly downregulated relative to the normal samples (P < 0.05). Moreover, in the GSE30219 dataset (Fig. 2H), except for BMP6, the expression levels of all other 17 IEGs (CAT, CD79, CD83, CSF3, CXCL12, CXCL2, CXCR2, HLA-E, HSPD1, HSPE1, IL6, PPARG, S1PR1, S1PR4, SELE, TIMP1, TLR4) were significantly different between the LUAD and normal groups (P < 0.05); among them, CD79A, HSPD1, HSPE1, TIMP1 were significantly upregulated, whereas CAT, CD83, CSF3, CXCL12, CXCL2, CXCR2, HLA-E, IL6, PPARG, S1PR1, S1PR4, SELE, TLR4 was significantly downregulated (P < 0.05). The differential analysis results of the GSE10072 dataset (Fig. 2I) showed that the expression of all 18 IEGs was significantly different between the LUAD and normal groups (P < 0.05) genes CD79A, HSPD1, HSPE1, TIMP1 were significantly upregulated (P < 0.05) in the LUAD group, whereas genes BMP6, CAT, CD83, CSF3, CXCL12, CXCL2, CXCR2, HLA-E, IL6, PPARG, S1PR1, S1PR4, SELE, TLR4 were significantly downregulated (P < 0.05). These results suggest that the 18 identified IEGs may play an important role in LUAD. Mutation analysis of IEGs in patients with LUAD To analyze the somatic mutations in the 18 IEGs of the patients with LUAD included in TCGA-LUAD, we compiled the mutation analysis results corresponding to the 18 IEGs. The results indicated that the main types of somatic cell mutations in the TCGA-LUAD dataset were missense and nonsense mutations; however, frame-shift insertion (Frame-Shift-Ins), in-frame insertion (In-Frame-Ins), and splice-site mutations were also observed. These mutations mainly involved single nucleotides (leading to SNPs) or the insertion (INS) or deletion (DEL) of short DNA fragments. Moreover, C > A was the most common single nucleotide variation in patients with LUAD, followed by C > T and C > G (Fig. 3A). As shown in Fig. 3B, somatic mutations existed in 17 genes(BMP6, CAT, CD79A, CD83, CXCL12, CXCL2, CXCR2, HLA-E, HSPD1, HSPE1, IL6, PPARG, S1PR1, S1PR4, SELE, TIMP1, TLR4) of the 18 IEGs in the TCGA-LUAD dataset. Moreover, among the 148 different somatic mutations found in patients with LUAD, 125 (84.46%) were found in these 17 IEGs. Missense mutations were the most abundant mutations among the 17 IEGs, and TLR4 had the highest mutation rate, accounting for 45% of the mutations in patients with LUAD. Mutation analysis of IRDEGs in LUAD. A. The mutation status of IRDEGs in LUAD. B. Detailed display of IRDEGs mutations in LUAD. C. IRDEGs copy number change frequency display. D. Display of copy number variation of IRDEGs. E. Chromosomal map of IRDEGs. IRDEGs, Immunomodulatory-related differentially expressed genes. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma Subsequently, by analyzing the copy number variation of the IEGs in LUAD (Fig. 3D), it was found that most samples had copy number changes, mainly reductions. However, some genes showed high copy numbers; for example, gene S had a high copy number amplification frequency (40%–60%), and there were almost no samples with missing copies. Some genes had a higher copy number change frequency than others; for example, genes SELE and IL6 had the highest copy number change frequencies (Fig. 3C). We used the RCircos package to annotate the positions of the 18 IEGs on human chromosomes (Fig. 3E). As shown in Fig. 3E, the 18 IEGs were mainly distributed on chromosomes 1, 2, 3, 4, 6, 7, 9, 10, 11, 17, 19, and X, being chromosomes 2 and 6 those with the highest number of IEGs: three IEGs each. These data suggest that those IEGs located on the same chromosome may have a relatively close connection at the genomic level. Gene Ontology function and KEGG pathway enrichment analyses of IEGs To investigate the relationship between the biological processes, molecular functions, cell components, and biological pathways of the 18 IEGs and LUAD, we first conducted GO and KEGG pathway enrichment analyses, and the results are presented in a bar chart (Fig. 4A). Subsequently, the relationship between IEGs and the results of GO and KEGG enrichment analyses were used to construct a network diagram (Fig. 4B). The line in the diagram indicates the corresponding relationship between the numerator of the network matrix and item annotation. The larger the node, the more numerators an item contains. The selection criteria for enrichment items were P adjust < 0.05 and FDR (q value) < 0.25, which were considered statistically significant. The results showed that the 18 IEGs in the LUAD group were mainly enriched in BP —including positive regulation of interleukin-10 production, regulation of cell–cell adhesion, and interleukin-10 production—, CC —including the external side of the plasma membrane, clathrin-coated pit, and membrane raft—, and MF —including H1 activity, growth factor activity, and receptor ligand activity—. KEGG pathway enrichment analysis showed that the 18 IEGs were significantly enriched in a variety of pathways, such as lipid and atherosclerosis, malaria, and legionellosis pathways (Table 2). GO and KEGG analysis of IRDEGs. A Histogram display of GO and KEGG analysis results of IRDEGs. B Network diagram display of the relationship between IRDEGs and their GO and KEGG analysis results. Yellow nodes represent specific functional pathways, and green nodes represent genes. C The results of IRDEGs joint logFC GO and KEGG analysis results are shown in circle diagrams. D Chord diagram display of GO and KEGG analysis results of IRDEGs combined with logFC. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. IRDEGs, Immunomodulatory-related differentially expressed genes. GO, Gene Ontology; BP, biological process; CC, cellular component; MF, molecular function. KEGG, Kyoto Encyclopedia of Genes and Genomes. The screening criteria for GO and KEGG enrichment items were P. adjust < 0.05 and FDR value (q. value) < 0.25 Table 2GO and KEGG enrichment analysis results of IRDEGsOntologyIDDescriptionGeneRatioBgRatiopvaluep.adjustBPGO:0032733positive regulation of interleukin-10 production4/1841/188005.83e-087.65e-05BPGO:0022407regulation of cell–cell adhesion7/18456/188001.19e-077.82e-05BPGO:0032613interleukin-10 production4/1863/188003.38e-070.0001BPGO:0032653regulation of interleukin-10 production4/1863/188003.38e-070.0001BPGO:0032496response to lipopolysaccharide6/18333/188004.58e-070.0001CCGO:0009897external side of plasma membrane8/18455/195942.84e-092.04e-07CCGO:0005905clathrin-coated pit2/1870/195940.00190.0562CCGO:0045121membrane raft3/18326/195940.00310.0562CCGO:0098857membrane microdomain3/18327/195940.00310.0562CCGO:0005769early endosome3/18389/195940.00510.0731MFGO:0005125cytokine activity6/18235/184106.63e-086.76e-06MFGO:0008083growth factor activity5/18162/184103.87e-071.98e-05MFGO:0048018receptor ligand activity6/18489/184104.82e-060.0001MFGO:0030546signaling receptor activator activity6/18496/184105.23e-060.0001MFGO:0045125bioactive lipid receptor activity2/1816/184100.00010.0020KEGGhsa05417Lipid and atherosclerosis6/16215/81642e-060.0001KEGGhsa05144Malaria4/1650/81642.15e-060.0001KEGGhsa05134Legionellosis4/1657/81643.65e-060.0001KEGGhsa04060Cytokine-cytokine receptor interaction6/16295/81641.25e-050.0003KEGGhsa05323Rheumatoid arthritis4/1693/81642.59e-050.0005GO, Gene Ontology. BP, biological process. CC, cellular component. MF, molecular function. KEGG, Kyoto Encyclopedia of Genes and Genomes. IRDEGs, Immunomodulatory-related differentially expressed genes GO and KEGG enrichment analysis results of IRDEGs GO, Gene Ontology. BP, biological process. CC, cellular component. MF, molecular function. KEGG, Kyoto Encyclopedia of Genes and Genomes. IRDEGs, Immunomodulatory-related differentially expressed genes Subsequently, we conducted GO and KEGG enrichment analyses combined with logFC analysis for the 18 IEGs. Based on the enrichment analyses and the logFC value of the IEGs from the differential analysis results of the TCGA-LUAD dataset, the z-score of each gene was calculated. These results were used to construct loop (Fig. 4C) and chord (Fig. 4D) diagrams that show that the 18 IEGs are mainly involved in the function of genes on the external side of the plasma membrane (CC) and the regulation of cell–cell adhesion (BP). GSEA between LUAD and normal groups To elucidate the impact of differential gene expression levels between the LUAD and normal groups on the occurrence of LUAD, we used GSEA to determine the association between these genes and the participating BPs, CCs, and MFs in TCGA-LUAD. P adjust < 0.05 and FDR (q value) < 0.25 were used as remarkable enrichment screening criteria, and a mountain map was constructed based on the enrichment results (Fig. 5A). The results suggested that there are statistically significant differences in gene expression between the LUAD and normal groups in the TCGA-LUAD dataset; these differences are more prominent in the resistin as a regulator of inflammation (Fig. 5B), cellular response to hypoxia (Fig. 5C), oxidative stress-induced senescence (Fig. 5D), and immunoregulatory interactions between a lymphoid and a non-lymphoid cell (Fig. 5E) pathways (Table 3). GSEA between LUAD and Normal groups in TCGA- LUAD dataset. A GSEA of genes between LUAD and Normal groups in the TCGA-LUAD data set shows the mountain map of the main four biological characteristics. B–E Genes between LUAD and Normal groups in TCGA-LUAD dataset are significantly enriched in Resistin As A Regulator Of Inflammation (B), Cellular Response To Hypoxia (C), Oxidative Stress Induced Senescence (D), Immunoregulatory Interactions Between A Lymphoid And A Non-lymphoid Cell (E) pathway. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. DEGs, differentially expressed genes. GSEA: Gene Set Enrichment Analysis. The significant enrichment screening criteria of GSEA were P.adjust < 0.05 and FDR value ( q.value) < 0.25 Table 3GSEA enrichment analysis resultsIDsetSizeenrichmentScoreNESpvaluep.adjustqvalueREACTOME_CELL_CYCLE_CHECKPOINTS2900.66073552.5698631e−109.69e−096.62e−09REACTOME_G2_M_CHECKPOINTS1660.67662122.4937831e−109.69e−096.62e−09REACTOME_DNA_REPLICATION1870.64957012.4356091e−109.69e−096.62e−09WP_RETINOBLASTOMA_GENE_IN_CANCER870.71005822.3777911e−109.69e−096.62e−09REACTOME_SYNTHESIS_OF_DNA1210.67418112.3701781e−109.69e−096.62e−09REACTOME_TRANSLATION2930.60520452.3572711e−109.69e−096.62e−09REACTOME_DNA_REPLICATION_PRE_INITIATION1590.63194602.3190211e−109.69e−096.62e−09REACTOME_MITOTIC_SPINDLE_CHECKPOINT1110.66443352.3150971e−109.69e−096.62e−09REACTOME_S_PHASE1630.62351672.3000381e−109.69e−096.62e−09REACTOME_RRNA_PROCESSING2030.60854162.2916151e−109.69e−096.62e−09REACTOME_IMMUNOREGULATORY_INTERACTIONS_BETWEEN_A_LYMPHOID_AND_A_NON_LYMPHOID_CELL128−0.542766834−1.8959353741.04315E−050.0001704250.000116368REACTOME_OXIDATIVE_STRESS_INDUCED_SENESCENCE1230.4632256961.6382634630.0003056070.0025919980.001769844WP_RESISTIN_AS_A_REGULATOR_OF_INFLAMMATION33−0.602976816−1.6804632710.0053902760.0257855630.017606657REACTOME_CELLULAR_RESPONSE_TO_HYPOXIA750.4579048931.512933650.0097659360.0412066860.028136364GSEA,Gene Set Enrichment Analysis GSEA enrichment analysis results GSEA,Gene Set Enrichment Analysis Determination of LUAD disease subtypes To inquire into the differential expression of IEGs in the LUAD group of the TCGA-LUAD dataset and aiming to identify different LUAD disease subtypes, we performed a consensus clustering by using ConsensusClusterPlus, of the 18 IEGs and identified four subtypes of LUAD disease (namely cluster1, cluster2, cluster3, and cluster4; Fig. 6A). LUAD disease cluster1 consisted of 140 samples, whereas cluster2 consisted of 132 samples, cluster3 of 183, and cluster4 of 84. pheatmap was then used to generate a heatmap of the 18 IEGs across the four LUAD disease subtypes identified in the TCGA-LUAD dataset (Fig. 6B). A Construction of related disease subtypes of LUAD. Consensus clustering (K = 4) results plot of LUAD disease. B Complex numerical heatmap of IRDEGs in different LUAD disease subtypes. C Group comparison chart of IRDEGs in different LUAD disease subtypes. The symbol ns is equivalent to P ≥ 0.05, no statistical significance; the symbol * is equivalent to P < 0.05; the symbol ** is equivalent to P < 0.01; the symbol *** is equivalent to P < 0.001. LUAD, Lung adenocarcinoma. IRDEGs, Immunomodulatory-related differentially expressed genes Thereafter, we used the Mann–Whitney U test to analyze the discrepancy in the expression levels of the 18 IEGs among the four LUAD disease subtypes in the TCGA-LUAD dataset and visualized the results through a grouping comparison chart (Fig. 6C). As shown in Fig. 6C, there were statistically significant differences (P < 0.05) in the expression of the 18 IEGs among the four LUAD disease subtypes in the TCGA-LUAD dataset. Analysis of immune infiltration and immunotherapy base on LUAD disease subtypes After sorting the expression profile data of the TCGA-LUAD dataset, we used the ssGSEA algorithm to calculate the infiltration of 28 types of immune cells and plotted a correlation heatmap (Fig. 7A) between the immune cells of the ssGSEA algorithm and the 18 IEGs. Subsequently, we drew a group comparison diagram (Fig. 7B) of the infiltrated immune cells in the four LUAD disease subtypes. The results indicated that there was a statistically significant difference in the presence of 26 immune cells among the four LUAD disease subtypes, CD56dim natural killer and memory B cells the exceptions. As shown in Fig. 7A, activated dendritic cells were the most significantly differently infiltrated immune cells in all four disease subtypes (P < 0.001). We then plotted a lollipop chart of the correlation between activated dendritic cells and the 18 IEGs (Fig. 7C). Analysis of immune infiltration (ssGSEA) and immunotherapy based on disease subtype. A Correlation heat map of immune cells and IRDEGs in the ssGSEA algorithm in the dataset TCGA- LUAD. B The boxplot of the difference of immune cells in different LUAD disease subtypes by ssGSEA algorithm. C Correlation lollipop graph between activated dendritic cell and 18 IRDEGs. D–F IPS_CTLA4_NEG_PD1_NEG (D), IPS_CTLA4_NEG_PD1_POS (E), IPS_CTLA4_POS_PD1_NEG (F), IPS_CTLA4_POS_PD1_POS (G) group comparison chart among different disease subtypes. The symbol ns is equivalent to P ≥ 0.05, which is not statistically significant; the symbol * is equivalent to P < 0.05, which is statistically significant; the symbol ** is equivalent to P < 0.01, which is highly statistically significant; the symbol *** is equivalent to P < 0.001, which is extremely statistically significant. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. IRDEGs, Immunomodulatory-related differentially expressed genes To further analyze the prediction ability of the disease subgroup models for immunotherapy, we downloaded the LUAD-related immunogenicity scores (i.e,. the IPSs) from the TCIA database and drew difference boxplots of TCGA-LUAD patients with different LUAD disease subtypes and IPSs using ggplot in R (IPS_CTLA4_NEG_PD1_NEG, Fig. 7D; IPS_CTLA4_NEG_PD1_POS, Fig. 7E; IPS_CTLA4_POS_PD1_NEG, Fig. 7F; and IPS_CTLA4_POS_PD1_POS, Fig. 7G). The results showed significant differences among the four IPS subtypes, suggesting that the different LUAD subtypes can predict the prognosis of immunotherapy with PD1 and CTLA4. Constructing the Is Based on the expression of the 18 IEGs in the TCGA-LUAD dataset, we obtained the Is of each patient in the dataset using the ssGSEA algorithm (each Is represents the immunomodulatory level of each patient). We then plotted the diagnostic ROC curve of the LUAD and normal groups in the TCGA-LUAD dataset using these scores (Fig. 8A). These Is were highly accurate for diagnosing patients and classifying them as from the LUAD or normal groups (AUC = 0.971; Fig. 8A). We then used the Mann–Whitney U test to analyze the divergence between the scores of the LUAD and normal groups in TCGA-LUAD, and the results showed statistically significant differences (P < 0.001; Fig. 8B). Construction of immune regulation score. A, B The ROC curve display (A) and the group comparison graph display (B) of the immune regulation score between the LUAD group and the Normal group in the TCGA-LUAD dataset. C, D Is were shown in the group comparison charts between High-Is group and low-Is group in the TCGA-LUAD data set (C) and the GSE3141 data set (Fig. 8D). P ≥ 0.05, no statistical significance; P < 0.05, statistically significant; P < 0.01, highly statistically significant; P < 0.001, extremely statistically significant. The closer the AUC is to 1 in the ROC curve, the better the diagnostic effect. When AUC is 0.5–0.7, it has low accuracy, when AUC is 0.7–0.9, it has certain accuracy, and when AUC is above 0.9, it has high accuracy. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. ROC, receiver operating characteristic curve. IRDEGs, Immunomodulatory-related differentially expressed genes. Is, Immunomodulatory score Similarly, we obtained the Is for each patient with LUAD from the GSE3141 dataset using the ssGSEA algorithm. According to the median Is, TCGA-LUAD and GSE3141 dataset samples were divided into high- and low-Is groups. We then used the Mann–Whitney U test to analyze the difference in scores between the two groups in the TCGA-LUAD (Fig. 8C) and GSE3141 (Fig. 8D) datasets. The evidence indicated that in both the TCGA-LUAD and GSE3141 datasets, the scores calculated by us were statistically significantly different (P < 0.001). To further evaluate the model’s predictive performance, detailed ROC-curve parameters are provided (Table S2). Taking the feature “Is” as an example, the optimal cutoff is 0.7329. At this threshold, the model achieves a sensitivity of 0.9661, correctly identifying 96.61% of positive cases, and a specificity of 0.8664, correctly identifying 86.64% of negative cases. The area under the receiver-operating characteristic curve (AUC) is 0.9706, indicating exceptional discriminatory ability. Moreover, the 95% confidence interval for AUC is 0.9560–0.9852, further confirming the stability and reliability of the model’s predictions. Correlation analysis of IEGs in the LUAD cancer group in TCGA-LUAD We used the Spearman algorithm to analyze the correlation between the expression levels of the 18 IEGs in patients with LUAD in TCGA-LUAD and visualized the results through a correlation heatmap (Fig. 9C). We found a positive correlation between the expression levels of the 18 IEGs in the TCGA-LUAD cancer group. We then determined the correlation between the top 2 pairs of genes with the highest and lowest correlations in the correlation heatmap through a correlation scatter plot (Fig. 9A, B, D, E). Among the 18 IEGs, the expression levels of HSPE1 and HSPD1 (r = 0.663, P < 0.001; Fig. 9A) and of TLR4 and CXCL12 (r = 0.615, P < 0.001; Fig. 9B) in the LUAD group in TCGA-ESCC showed remarkable positive correlations, whereas the expression levels of HSPD1 and CD83 (r = − 0.364, P < 0.001; Fig. 9D) and of S1PR1 and HSPE1 (r = − 0.351, P < 0.001; Fig. 9E) in the LUAD cancer group in TCGA-ESCC showed significant negative correlations. Correlation analysis of IRDEGs in cancer group of TCGA-LUAD dataset. A Scatter plot showing the correlation between gene HSPE1 and gene HSPD1. B Scatter plot showing the correlation between gene TLR4 and gene CXCL12. C The correlation heat map results of IRDEGs in the cancer group samples of the TCGA-LUAD dataset. D Scatter plot display of the correlation between gene HSPD1 and gene CD83. E Scatter plot showing the correlation between gene S1PR1 and gene HSPE1. The symbol ns is equivalent to P ≥ 0.05, which is not statistically significant; the symbol * is equivalent to P < 0.05, which is statistically significant; the symbol ** is equivalent to P < 0.01, which is highly statistically significant; the symbol *** is equivalent to P < 0.001, which is highly statistically significant. The correlation coefficient (r) in the correlation scatter plot is strong correlation if the absolute value is above 0.8; moderate correlation if the absolute value is 0.5–0.8; weak correlation if the absolute value is 0.3–0.5; weak or no correlation if the absolute value is below 0.3. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. IRDEGs, Immunomodulatory-related differentially expressed genes GSVA between high- and low-Is groups To probe the difference in the hallmark gene set between the high- and low-Is groups in TCGA-LUAD, we conducted a GSVA (Fig. 10A–D, Tables 4 and 5) on the genes from the two groups in the TCGA-LUAD (Fig. 10A, B) and GSE3141 (Fig. 10C, D) datasets. GSVA gene set enrichment analysis between high and low Is groups. A, B The heat map (A) of the GSVA results of genes between High-Is group and low-Is group in the TCGA-LUAD dataset, and the group comparison map (B) shows. C, D The heat map (C) of the GSVA results of genes between High-Is group and low-Is group in the GSE3141 dataset, and the group comparison map (D) shows. The symbol ns is equivalent to P ≥ 0.05, which is not statistically significant; the symbol * is equivalent to P < 0.05, which is statistically significant; the symbol ** is equivalent to P < 0.01, which is highly statistically significant; the symbol *** is equivalent to P < 0.001, which is highly statistically significant. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. Is, Immunomodulatory score. GSVA, Gene Set Variation Analysis Table 4GSVA analysis of dataset TCGA−LUADontologylogFCAveExprtP.Valueadj.PHALLMARK_INFLAMMATORY_RESPONSE−0.358500645−0.014153817−16.092465887.35E−493.67E−47HALLMARK_IL2_STAT5_SIGNALING−0.28954932−0.022887388−15.452513699.91E−462.48E−44HALLMARK_KRAS_SIGNALING_UP−0.296529142−0.007137079−15.369538822.50E−454.17E−44HALLMARK_TNFA_SIGNALING_VIA_NFKB−0.352089552−0.028585249−15.015452211.27E−431.58E−42HALLMARK_IL6_JAK_STAT3_SIGNALING−0.350553861−0.008533063−14.91712133.74E−433.74E−42HALLMARK_COMPLEMENT−0.300146666−0.011978081−14.844951288.25E−436.87E−42HALLMARK_ALLOGRAFT_REJECTION−0.329271787−0.006637719−12.77395882.69E−331.92E−32HALLMARK_COAGULATION−0.2094715590.013957864−10.176072671.41E−228.78E−22HALLMARK_ANGIOGENESIS−0.243442946−0.009937824−9.2457199553.82E−192.12E−18HALLMARK_APICAL_SURFACE−0.201507487−0.003907275−9.1094184311.16E−185.82E−18HALLMARK_MYOGENESIS−0.183942461−0.009716053−8.9512999584.16E−181.89E−17HALLMARK_APOPTOSIS−0.175422276−0.016480617−8.7796829191.63E−176.79E−17HALLMARK_DNA_REPAIR0.201386097−0.0198835828.7676683451.79E−176.89E−17HALLMARK_INTERFERON_GAMMA_RESPONSE−0.227883604−0.005559178−8.1980147791.43E−155.12E−15HALLMARK_MYC_TARGETS_V20.249362916−0.023078727.4572584513.03E−131.01E−12HALLMARK_E2F_TARGETS0.249186637−0.0205430637.3195521357.85E−132.45E−12HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION−0.208772889−0.022755118−7.243972271.31E−123.86E−12HALLMARK_G2M_CHECKPOINT0.223670214−0.020980097.0910261213.68E−121.02E−11HALLMARK_GLYCOLYSIS0.161302997−0.0042107667.0059949666.48E−121.71E−11HALLMARK_UV_RESPONSE_DN−0.161177909−0.02938735−6.9251704421.10E−112.76E−11HALLMARK_HYPOXIA−0.139951969−0.01801204−6.6920921384.97E−111.18E−10HALLMARK_HEME_METABOLISM−0.115062196−0.025922401−6.6276766387.48E−111.70E−10HALLMARK_MYC_TARGETS_V10.211081313−0.0231287686.5579386691.16E−102.52E−10HALLMARK_TGF_BETA_SIGNALING−0.145713426−0.023932722−5.7967509961.08E−082.26E−08HALLMARK_XENOBIOTIC_METABOLISM−0.107389322−0.008389888−5.5956374523.32E−086.63E−08HALLMARK_MITOTIC_SPINDLE0.128488421−0.0326045695.5315621284.71E−089.05E−08HALLMARK_UNFOLDED_PROTEIN_RESPONSE0.13588733−0.0146587925.3184445011.47E−072.72E−07HALLMARK_ANDROGEN_RESPONSE−0.109283164−0.044394305−5.1451578723.60E−076.44E−07HALLMARK_BILE_ACID_METABOLISM−0.102051964−0.005600337−5.1088634184.33E−077.47E−07HALLMARK_APICAL_JUNCTION−0.105191926−0.002694109−4.972911088.57E−071.43E−06HALLMARK_P53_PATHWAY−0.09356083−0.016559566−4.833751741.70E−062.74E−06HALLMARK_PANCREAS_BETA_CELLS−0.1004745870.057084288−4.6308903174.45E−066.95E−06HALLMARK_HEDGEHOG_SIGNALING−0.1163096430.001322139−4.5559556786.30E−069.54E−06HALLMARK_KRAS_SIGNALING_DN−0.0732321480.029135014−4.4699992079.32E−061.37E−05HALLMARK_INTERFERON_ALPHA_RESPONSE−0.117904918−0.006956008−3.750451110.0001932550.000276079HALLMARK_SPERMATOGENESIS0.0648237740.0019943983.6925171940.0002418630.00033592HALLMARK_ESTROGEN_RESPONSE_LATE−0.069363306−0.027214186−3.670828580.0002628550.000355209HALLMARK_CHOLESTEROL_HOMEOSTASIS-0.085839525−0.034118568−3.5892784720.0003581120.0004712HALLMARK_MTORC1_SIGNALING0.093357879−0.0249524383.5684332480.0003872040.000496415HALLMARK_ESTROGEN_RESPONSE_EARLY−0.06963045−0.024344347−3.5499353580.0004148590.000518574HALLMARK_OXIDATIVE_PHOSPHORYLATION0.104605355−0.0391330553.4681240660.0005608390.000683949HALLMARK_WNT_BETA_CATENIN_SIGNALING−0.074637646−0.005328006−3.1832838280.0015299130.001821325HALLMARK_UV_RESPONSE_UP−0.059979288−0.022698464−3.0742169690.0022044140.002563272HALLMARK_PROTEIN_SECRETION0.06228467−0.0386396872.5557036410.0108375530.012315401HALLMARK_PEROXISOME0.049067268−0.0192292.4817633380.0133403410.014822601TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. GSVA, Gene Set Variation Analysis GSVA analysis of dataset TCGA−LUAD TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. GSVA, Gene Set Variation Analysis Table 5GSVA analysis of dataset GSE3141ontologylogFCAveExprtP.Valueadj.PHALLMARK_COMPLEMENT−0.388960070.011959233−9.5382679381.40E−167.02E−15HALLMARK_KRAS_SIGNALING_UP−0.3666570960.008564602−9.3745241063.51E−167.81E−15HALLMARK_COAGULATION−0.3847676820.010890262−9.3229610284.68E−167.81E−15HALLMARK_IL6_JAK_STAT3_SIGNALING−0.4177370740.01738235−8.5955089882.65E−143.31E−13HALLMARK_INFLAMMATORY_RESPONSE−0.3737303630.019805882−8.3951744477.95E−147.95E−13HALLMARK_ALLOGRAFT_REJECTION−0.4097611730.000661659−8.1071231753.81E−133.18E−12HALLMARK_IL2_STAT5_SIGNALING−0.3165866170.012454832−8.0118563396.38E−134.55E−12HALLMARK_APOPTOSIS−0.2812434990.009444107−7.0407356121.08E−106.74E−10HALLMARK_INTERFERON_GAMMA_RESPONSE−0.3756922510.022365082−6.5799725221.12E−096.24E−09HALLMARK_MYC_TARGETS_V20.4061561016.86E−056.4396711822.26E−091.13E−08HALLMARK_TNFA_SIGNALING_VIA_NFKB−0.3290289020.007025788−6.3780143933.07E−091.40E−08HALLMARK_ANGIOGENESIS−0.3423671360.011071669−6.3245568133.99E−091.66E−08HALLMARK_E2F_TARGETS0.408273859−0.004742466.1816301418.04E−093.09E−08HALLMARK_G2M_CHECKPOINT0.3678229160.001326476.0698231031.38E−084.93E−08HALLMARK_MYOGENESIS−0.2317094590.005007086−5.7632081145.94E−081.98E−07HALLMARK_MYC_TARGETS_V10.3547508480.0074354675.6711283549.14E−082.86E−07HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION−0.3108290420.008386173−5.5082468141.94E−075.70E−07HALLMARK_SPERMATOGENESIS0.1889287390.0103687654.9745905472.08E−065.79E−06HALLMARK_INTERFERON_ALPHA_RESPONSE−0.3194285690.031024608−4.8023152054.35E−061.14E−05HALLMARK_UV_RESPONSE_DN−0.2085204390.000592696−4.646789058.33E−062.02E−05HALLMARK_GLYCOLYSIS0.2020751890.0295433414.6426739948.48E−062.02E−05HALLMARK_DNA_REPAIR0.2064789330.0073999984.6062676349.85E−062.24E−05HALLMARK_HEME_METABOLISM−0.1664690930.001495989−4.546649471.26E−052.73E−05HALLMARK_APICAL_JUNCTION−0.1790403870.015549142−4.2485421264.14E−058.62E−05HALLMARK_MTORC1_SIGNALING0.2018254830.018742893.8348226830.0001971150.00039423HALLMARK_TGF_BETA_SIGNALING−0.20872250.011379384−3.677748290.0003464680.000666284HALLMARK_OXIDATIVE_PHOSPHORYLATION0.1740285090.0097377353.2567467810.0014464310.002678575HALLMARK_MITOTIC_SPINDLE0.1390081870.0070747223.1848121090.0018237030.003256612HALLMARK_XENOBIOTIC_METABOLISM−0.1199085210.020577983−2.9894528070.0033584980.005790514HALLMARK_BILE_ACID_METABOLISM−0.112649336−0.004232805−2.7155799060.0075404090.012567349HALLMARK_PANCREAS_BETA_CELLS−0.116874980.002309907−2.5200964040.0129756810.020928518HALLMARK_ANDROGEN_RESPONSE−0.105850040.008596344−2.4447148880.0158725750.024800898HALLMARK_UNFOLDED_PROTEIN_RESPONSE0.1075884630.0226075732.2672402480.0250708940.037986203HALLMARK_APICAL_SURFACE−0.1062760140.007837263−2.2203613340.0281730440.041430947GSVA,Gene Set Variation Analysis GSVA analysis of dataset GSE3141 GSVA,Gene Set Variation Analysis The GSVA results identified 45 hallmark gene sets, such as allograft rejection, androgen response, angiogenesis, and apical junction, with significant differences between the high- and low-Is groups in TCGA-LUAD (see Table 4). In light of these results, we analyzed the differential expression of the 45 hallmark pathways between the high- and low-Is groups and drew a heatmap to visualize the results (Fig. 10A). The Mann–Whitney U test was then used to analyze the grouping differences of the 45 hallmark pathways between the high- and low-Is groups in the TCGA-LUAD dataset, and the outcomes were displayed using a group comparison chart (Fig. 10B). The GSVA results identified 34 hallmark gene sets, such as allograft rejection, androgen response, angiogenesis, and apical junction, with significant differences between the high- and low-Is groups in GSE3141 (Table 5). We then analyzed the differential expression of the 34 hallmark pathways between the high- and low-Is groups in GSE3141 and drew a heatmap to visualize the results (Fig. 10C). We analyzed the expression differences of the 34 hallmark pathways between the high- and low-Is groups in GSE3141 using the Mann–Whitney U test and constructed group comparison graphs for presenting the results (Fig. 10D). Construction of an IEG-based prognostic model To confirm the prognostic value of the 18 IEGs in TCGA-LUAD, we constructed a prognostic model based on these IEGs using LASSO regression analysis (Fig. 11A). After visualizing the LASSO regression results, we obtained a LASSO variable trajectory map (Fig. 11C). Figure 11 shows that there are six key genes in the IEG-based prognostic model, namely TIMP1, CD79A, CAT, PPARG, TLR4, HSPD1. Construction of IRDEGs prognosis model. A LASSO regression prognostic model diagram of IRDEGs. B, C Variable trajectory plot (B) and risk factor plot (C) of the IRDEGs prognostic model. D–G Prognostic survival KM curves of genes CAT, CD79A, HSPD1, TLR4. TCGA, The cancer genome atlas. LUAD, Lung adenocarcinoma. IRDEGs, Immunomodulatory-related differentially expressed genes. LASSO, Least absolute shrinkage and selection operator We then displayed the sample groupings in the constructed IEG-based prognostic model using a risk factor map (Fig. 11B). This risk factor map consists of three parts: (1) Risk grouping: the risk scores predicted by the IEG-based prognostic model for the dataset samples were grouped by the median; (2) survival outcome: the survival time and survival outcome of the clinical samples in TCGA-LUAD were used to generate a dot plot; and (3) heatmap: the expression of the key prognostic genes in the IEG-based prognostic model is visualized through prognostic model samples. We then drew individual prognostic survival KM curves for the six prognostic key genes, and those genes with statistically significant different expression levels (P < 0.05) were considered relevant. From these analyses, four genes were considered key prognostic genes (Fig. 11D–G). The expression levels of genes CAT, CD79A, HSPD1, and TLR4 in TCGA-LUAD showed a certain relationship with the prognosis of patients with LUAD: CAT (P = 0.005, Fig. 11D), CD79A (P = 0.017, Fig. 11E), HSPD1(P = 0.023, Fig. 11F), TLR4(P = 0.006, Fig. 11G). Prognostic performance of the key genes in the prognostic model To further validate the IEG-based prognostic model, we conducted a statistical analysis of the clinical data of patients with LUAD in TCGA-LUAD (Table 6). Subsequently, we used univariate and multivariate Cox regression analyses to determine the correlation between the expression levels of six key prognostic genes and five clinical variables (T stage, N stage, M stage, age, and sex assigned at birth) and the clinical prognosis of patients with LUAD. First, the five clinical variables and the expression levels of the six key genes were analyzed using univariate Cox regression, and those variables satisfying P < 0.10 in the univariate analysis results were further used in a multivariate Cox regression analysis to construct a multivariate Cox regression model. The results showed that the clinical variables T, N, and M stages, and the expression of the six key prognostic genes were significantly correlated with the prognosis of patients with LUAD (P < 0.10, Table 7). We sorted the results of the multivariate Cox regression and presented them in a forest plot (Fig. 12A). Table 6Patient characteristics of LUAD patients in the TCGA-LUAD datasetsCharacteristicsoverallPathologic T stage, n (%)T1176 (32.8%)T2292 (54.5%)T349 (9.1%)T419 (3.5%)Pathologic N stage, n (%)N0350 (66.9%)N197 (18.5%)N274 (14.1%)N32 (0.4%)Pathologic M stage, n (%)M0365 (93.6%)M125 (6.4%)Gender, n (%)Female289 (53.6%)Male250 (46.4%)Age, n (%) < = 65257 (49.4%) > 65263 (50.6%) Patient characteristics of LUAD patients in the TCGA-LUAD datasets Table 7Cox regression to identify clinical features of dataset TCGA-LUADCharacteristicsTotal(N)Univariate analysisMultivariate analysisHazard ratio (95% CI)P valueHazard ratio (95% CI)P valuePathologic T stage527 < 0.001T1176ReferenceReferenceT22851.507 (1.059–2.146)0.0231.456 (0.920–2.305)0.109T3&T4663.095 (1.967–4.868) < 0.0012.447 (1.377–4.350)0.002Pathologic N stage514 < 0.001N0345ReferenceReferenceN1962.293 (1.632–3.221) < 0.0011.852 (1.248–2.747)0.002N2&N3732.993 (2.057–4.354) < 0.0012.395 (1.532–3.743) < 0.001Pathologic M stage3810.010M0356ReferenceReferenceM1252.176 (1.272–3.722)0.0051.550 (0.860–2.794)0.145Gender5300.570Female283ReferenceMale2471.087 (0.816–1.448)0.569Age5200.185 < = 65257Reference > 652631.216 (0.910–1.625)0.186TIMP15301.182 (1.013–1.379)0.0341.139 (0.942–1.379)0.179CD79A5300.870 (0.795–0.952)0.0020.900 (0.796–1.017)0.092CAT5300.721 (0.595–0.874) < 0.0010.775 (0.603–0.996)0.046PPARG5301.153 (0.996–1.336)0.0571.032 (0.866–1.229)0.727TLR45300.748 (0.601–0.932)0.0100.864 (0.654–1.141)0.304HSPD15301.469 (1.183–1.824) < 0.0011.137 (0.865–1.494)0.359 Cox regression to identify clinical features of dataset TCGA-LUAD Prognostic performance of the key genes prognostic model. A Forest plot of univariate Cox regression analysis of prognostic key genes combined with clinical variables. B Nomogram of multivariate Cox regression analysis of prognostic key genes combined with clinical variables. C–E The 1-year (C), 3-year (D), and 5-year (E) calibration curves of the Cox regression prognostic model nomogram analysis. F–H 1-year (F), 3-year (G), 5-year (H) DCA plots of the Cox regression prognostic model. DCA, decision curve analysis We then ran a nomogram analysis based on the results of the multivariate Cox regression analysis to judge the prognostic power of the model (Fig. 12B). The nomogram was based on the multivariate regression analysis by setting a certain scale to score and characterize the situation of each variable in the multivariate regression model and by calculating the total score to predict the probability of occurrence of an event. Additionally, we performed 1-year (Fig. 12C), 3-year (Fig. 12D), and 5-year (Fig. 12E) prognostic calibration analyses on the nomogram of the multivariate Cox regression model and constructed a calibration curve (Fig. 12C–E). The abscissa axis of the calibration curve represents the survival probability predicted by the model, and the ordinate axis represents the survival probability calculated from actual data. Different line and point colors represent the situations predicted by the model at different time points. The closer the lines of different colors to the ideal gray line, the better the prediction effect at that time point. We subsequently evaluated the clinical utility of the constructed Cox regression prognostic model using DCA at 1 (Fig. 12F), 3 (Fig. 12G), and 5 (Fig. 12H) years. The x and y axes in the DCA graph represent the probability thresholds and net returns, respectively. The results were measured by examining the x-value range in which the model line could be steadily higher than the all-positive and all-negative lines. The larger the x-value range, the better the model effect. The results showed that the clinical predictive effect of the Cox regression prognostic model was 5-year > 3-year > 1-year. Discussion LUAD is characterized by difficult early detection, high metastasis rate, susceptibility to chemotherapy, and radiation resistance. Despite the combination of various treatment methods, such as surgery, radiotherapy, chemotherapy, and targeted therapy, the 5-year survival rate of patients with LUAD remains poor [49]. Efforts should be made to improve the treatment outcomes and enhance the survival rate and quality of patients with LUAD. Genes involved in immune regulation play important roles in cancer research [14]. The immune system is a crucial mechanism for identifying and eliminating abnormal cells, and immune regulatory genes regulate the function of the immune system [15]. During cancer development, cancer cells can evade the monitoring and attack of the immune system, in which the abnormal expression or function of IEGs often plays a significant role. Several IRGs, including PD1, PD-L1, and CTLA4, have emerged as vital targets in cancer therapy. ICIs can restore the ability of the immune system to recognize and attack tumors by blocking the interaction between these immune regulatory molecules, thereby enhancing the survival rate and quality of life of some patients with cancer [8]. Furthermore, studies have revealed that polymorphisms are closely associated with cancer susceptibility and treatment efficacy. A comprehensive understanding of the mechanisms underlying the actions of these genes in cancer holds immense promise for advancing personalized treatments and precision medicine, thereby improving treatment outcomes in patients with cancer. This study innovatively focused on the correlation between IEGs and the diagnosis, treatment, and prognostic assessment of LUAD, contributing new ideas for the early diagnosis and treatment of this disease. In this study, we screened 18 IEGs between the LUAD and normal groups and analyzed the changes and functions of these genes in the progression of LUAD at the genomic level through mutation analysis. The results showed that mutations in IEGs were commonly present in the TCGA-LUAD dataset and that TLR4 had the highest mutation rate. TLR4 is involved in proinflammatory responses and plays a key role in innate immunity [50]. In vitro studies have shown that TLR4 is continuously expressed and upregulated in human LUAD cancer cells, and its expression level is significantly correlated with the production of immunosuppressive factors and angiogenic chemokines as well as with the resistance of cancer cells to apoptosis, thereby promoting the proliferation and survival of cancer cells [51, 52]. Recent studies have indicated that both cancer and immune cells, such as macrophages and dendritic cells within the TME, express TLR4. Activation of TLR4 in immune cells has been shown to inhibit tumor growth. However, the precise roles of TLR4 mutations in tumorigenesis and cancer progression remain unclear and warrant further investigation. The GO and KEGG analyses conducted in this study revealed that the 18 IEGs identified are primarily involved in extracellular substance synthesis, intercellular adhesion regulation, and IL-10 production, among other biological processes in LUAD. Tumor cell adhesion promotes tumor cell proliferation and metastasis and increases drug resistance [53, 54]. According to the GSEA enrichment analysis results, four pathways were significantly enriched in the LUAD and normal groups of the TCGA-LUAD dataset. Pathways related to oxidative stress-induced senescence and cellular responses to hypoxia were notably activated in the LUAD group. Conversely, the resistin as a regulator of inflammation and immunoregulatory interactions between lymphoid and non-lymphoid cell pathways were significantly inhibited. Resistin, a pro-inflammatory adipokine, has been demonstrated to play a critical role in tumor-related inflammation and immune regulation across multiple cancer types. Previous studies indicate that activation of the resistin signaling pathway is closely associated with chronic inflammation, immunosuppression, and tumor progression within the tumor microenvironment. Notably, our GSEA results revealed suppression of the "resistin as a regulator of inflammation" pathway in LUAD patients–contrasting with reported pathway activation in other malignancies. This suggests that LUAD may downregulate resistin-mediated inflammatory signaling to remodel the immunosuppressive TME, thereby facilitating immune evasion and tumor progression. This phenomenon may be attributed to tissue-specific immunosuppressive phenotypes, TME heterogeneity, or disease staging in LUAD [55]. Tumor cells usually have a high tolerance to hypoxic environments [56]. During hypoxia, within the tumor parenchyma, hypoxia inducible factor-1α (HIF-1α) is activated in cells to promote processes such as angiogenesis, glucose metabolism, and cell proliferation, thereby contributing to tumor cell survival in such a hypoxic environment [57]. Moreover, hypoxia in the TME induces oxidative stress in tumor cells [58]. Oxidative stress-induced cell senescence can cause cell cycle arrest, which helps limit the abnormal proliferation of tumor cells. This conclusion is in accordance with the results of our study, suggesting that the enrichment analysis results offer new insights into the pathogenesis of LUAD and may aid in the discovery of novel immunotherapeutic targets. We identified four LUAD disease subtypes using a consistent clustering method and assessed the 18 IEGs expression differences among these subtypes. As is well known, specific immune subtypes have unique immune characteristics and TME phenotypes in LUAD. The infiltration of 28 immune cells in the four subtypes was calculated, and the correlation between immune cells and the 18 IEGs was analyzed. The results indicated that in the TCGA-LUAD dataset, there were statistically significant differences in the infiltration of 26 immune cells among the different subtypes of LUAD disease. Notably, activated dendritic cells showed a high correlation with TLR4. Dendritic cells are considered the core component of the TME. After activation, they serve as specialized antigen-presenting cells that promote antitumor T cell responses [59, 60]. Studies have demonstrated that, in the high-risk LUAD groups, there is an increase in the ratio of T cells and resting dendritic and mast cells, whereas the proportion of activated dendritic and mast cells significantly decreases [3]. Consequently, the density and function of activated dendritic cells are considered key factors in the tumor immune response. Chen et al. indicated a correlation between TLR4 expression and the immune infiltration of CD8+ T, CD4+ T, and dendritic cells [61]. TLR4 promotes the maturation and infiltration of dendritic cells, thereby inhibiting the progression of NSCLC. Consistent with previous results, this study demonstrated an association between TLR4 and activated dendritic cells in LUAD. Therefore, we believe that TLR4 might be a new therapeutic target for LUAD with the potential to improve patient prognosis. It should be noted that the observed association between TLR4 and dendritic cells in this study is based solely on bioinformatics analyses; direct mechanistic validation through in vitro cellular assays or in vivo animal models is still lacking. Such experiments are essential to elucidate the precise mechanisms by which TLR4 regulates dendritic-cell function and shapes the immune microenvironment of lung adenocarcinoma. Therefore, future work will integrate cellular and animal models to systematically investigate the TLR4–dendritic-cell axis, aiming to clarify their interactive roles and underlying molecular mechanisms in lung adenocarcinoma and to advance the field from correlative findings to mechanistic insights. Next, we analyzed the Is differences between cancer and normal tissues. It has been found that Is decreases with an increase in tumor stage and has a good prognostic value in the overall survival of patients with colorectal cancer [62]. The calculation of Is in this study was based on the expression levels of the 18 IEGs. In the TCGA-LUAD dataset, the Is of the normal group apparently exceeded that of the LUAD group, which is consistent with previous research results. ROC curve analysis revealed that the Is had high diagnostic accuracy (AUC = 0.971). Subsequently, based on the median score, we categorized samples from the TCGA-LUAD and GSE31 datasets into high- and low-Is groups. We observed significant differences in the hallmark genes and pathways between the two groups, with DEGs primarily associated with allograft rejection, androgen response, and angiogenesis. Previous evidence indicates that androgens can modulate the activity of T cells and other immune cells, potentially enhancing immune escape in LUAD, thereby influencing the curative effect of immunotherapy [63]. Based on our results, we propose that Is could serve as a potential diagnostic marker for detecting and distinguishing patients with LUAD; however, further validation of the clinical application of Is is needed. We also screened six key genes through LASSO regression analysis to construct a prognostic model(TIMP1, CD79A, CAT, PPARG, TLR4, HSPD1). TIMP1 is an inhibitor of matrix metalloproteinases(MMPs) that can inhibit the invasion and metastasis of tumor cells [64]. However, many studies have shown that TIMP1 expression increases in some tumors, leading to poor prognosis [64, 65]. CD79A is involved in B cell development and the immune response. Studies have shown that high expression of CD79A is associated with the extensive infiltration of B cells into the TME. Zhang et al. independently validated the prognosis of CD79A and demonstrated that the high-expression group had a better prognosis, which is in line with the results of this study [66]. The CAT gene plays an antioxidant role in cells by encoding peroxidase, which effectively prevents cancer metastasis [67]. The overall survival of patients with LUAD and high CAT expression was significantly prolonged [68, 69]. In cancers, PPARG exhibits contradictory biological functions; PPARG overexpression advances the proliferation and survival of bladder cancer cells and plays an antiproliferative role in lung cancer [70, 71]. HSPD1, a member of the heat shock protein family, is crucial for the folding and assembly of proteins in mitochondria [72]. HSPD1 plays a vital role in the transformation, angiogenesis, and metastasis of cancer cells, and its high expression is associated with poor prognosis in patients with cancer [73]. Finally, we successfully constructed a multivariate Cox regression model incorporating T stage, N stage, M stage, and the six key prognostic genes identified here and the TCGA-LUAD dataset and confirmed that the model has good clinical utility in predicting the 5-year survival prognosis of patients with LUAD. Compared to existed panels mono-pathway models in the literature, our signature innovatively integrates three pivotal axes—immune regulation, stromal remodeling, and antioxidative stress response—thereby comprehensively capturing tumor microenvironment interaction networks. Furthermore, by dynamically incorporating TNM staging into the modeling framework, we achieved superior discriminatory power. This study still has limitations, such as a comparatively small sample size and possible selection bias. In addition, it was based solely on the TCGA-LUAD dataset for analysis and lacked support from external validation datasets. Moreover, this study lacks basic experimental evidence to demonstrate the mechanism of action of IEGs in LUAD. Therefore, further research on the functions and mechanisms of IEGs is needed. In this study, IEGs were analyzed using the TCGA-LUAD dataset, and the expression patterns of these genes in LUAD, as well as their correlation with the clinical features and prognosis of tumors, were determined. The potential value of IEGs in the prognostic evaluation of LUAD was assessed by constructing a multivariate regression model and drawing calibration and decision curves. This study suggests that these IEGs may be potential biomarkers for the prognostic evaluation of LUAD, providing new clues for a deeper understanding of the occurrence and progression of LUAD, and offering new ideas for its treatment and prognosis. Conclusion In conclusion, our study investigated the expression patterns of IEGs in LUAD, successfully classified LUAD into four subtypes based on the differential expression of IEGs, and constructed a prognostic model based on six key genes. Future investigations should prioritize additional experimental validation of immune-related genes in LUAD to potentially unveil novel immune checkpoints. Supplementary Information Additional file 1. Acknowledgements Not applicable. Abbreviations Lung adenocarcinoma Single-sample gene set enrichment analysis Immunomodulatory score Immunomodulatory-related genes Decision curve analysis Immune checkpoint inhibitor Tumor microenvironment Differentially expressed genes The immediate-early genes Tumor mutation burden Microsatellite instability Single nucleotide polymorphism Copy number variation Gene Ontology Biological process Molecular function Cellular component Kyoto Encyclopedia of Genes and Genomes Immunophenoscores Author contributions YLY: Analysis and Interpretation, Data Collection, Writing the Manuscript, Critical Revision, Statistical Analysis, Agreement to be Accountable, Approval of the Manuscript; CZM:Data Collection, Critical Revision, Agreement to be Accountable, Agreement to be Accountable, Approval of the Manuscript; XYF: Data Collection, Agreement to be Accountable; MT: Data Collection, Critical Revision, Statistical Analysis, Agreement to be Accountable, Approval of the Manuscript; YH: Conception and Design, Supervision, Critical Revision, Provide clinical data of patients, Agreement to be Accountable, Approval of the Manuscript; WDC: Conception and Design, Project administration, Supervision, Critical Revision, Agreement to be Accountable, Funding acquisition were performed, Approval of the Manuscript. Funding Beijing Municipal Public Welfare Development and Reform Pilot Project for Medical Research Institutes(PWD&RPP-MRI; Project No. JYY2023-14). Beijing Municipal Natural Science Foundation (grant number L234007). National Natural Science Foundation of China (grant numbers No. 82370087). Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate This study employed secondary de-identified genomic data from public repositories (TCGA/GEO). All publicly available data used in this study were sourced from previously published research. The study was exempt from ethical approval by the Beijing Chest Hospital Ethics Committee. All analyses were performed in accordance with the ethical principles of the Helsinki Declaration. Consent for publication All authors have approved the manuscript and agree with its submission to the journal. 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. Luyu Yang, Zhimin Cao and Xing Yuanfang have contributed equally to this work and should be considered as co-first authors. Teng Ma, Huan Ye and Dongchang Wang have contributed equally to this work and should be considered as Co-Corresponding authors. References Associated Data Supplementary Materials Additional file 1. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.