The GEO database was queried using the search terms “intracranial aneurysm” and “Homo sapiens” to retrieve publicly available gene expression profiles. The dataset GSE75436, generated on the GPL570 platform (Affymetrix Human Genome U133 Plus 2.0 Array), was designated as the training cohort. This dataset comprised gene expression array data from 10 intracranial aneurysm tissue specimens and 10 matched superficial temporal artery control samples. An independent validation cohort, GSE54083, was acquired from the GPL4133 platform (Agilent-014850 Whole Human Genome Microarray) and contained peripheral blood mononuclear cell (PBMC) expression profiles from 13 patients with intracranial aneurysm (disease group) and 13 healthy control individuals (control group). A comprehensive catalog of 3118 human immune-related genes was obtained from the ImmPort database.

Given the inherent differences in microarray platforms between the training set (GSE75436, Affymetrix) and the validation set (GSE54083, Agilent), preprocessing workflows were adapted to accommodate platform-specific characteristics while maintaining consistency in core analytical steps. All preprocessing procedures were executed using R software (version 4.3.1) for both the GSE75436 and GSE54083 datasets. The workflow comprised the following sequential steps:

Data Import: For the training set GSE75436, raw CEL files were imported utilizing the “affy” package in R to obtain the raw expression matrix, along with corresponding sample and gene expression information. For the validation set GSE54083, the processed data matrix provided by the platform was directly imported using the “GEOquery” package; this matrix had already undergone background correction via Agilent Feature Extraction software.

Background Adjustment: For the training set GSE75436, the Robust Multi-array Average (RMA) algorithm was applied to perform background correction, thereby minimizing noise introduced during chip hybridization. For the validation set GSE54083, background correction had been pre-performed by the data provider; therefore, this step was not repeated.

Quantile Normalization: The “limma” package was employed to conduct quantile normalization on the corrected or processed expression matrices for both datasets separately. This step standardized the distribution of gene expression values across all samples, mitigating systematic biases and ensuring inter-sample comparability [7].

Probe Annotation and Handling of Multi-Probe Mappings: Microarray probe identifiers were mapped to official gene symbols. Probes lacking a corresponding gene symbol or matching multiple genes were discarded. In instances where multiple probes corresponded to the same gene symbol, the median expression value across those probes was calculated and assigned as the representative expression value for that gene. Additionally, genes exhibiting expression levels within the lowest 25th percentile across all samples were filtered out to eliminate low-abundance transcripts that could introduce analytical noise.

Sample Quality Control: Principal component analysis (PCA) was utilized to detect potential outliers. Any sample displaying substantial deviation from other samples within the same experimental group was flagged and removed to uphold data integrity. Following PCA-based quality assessment, no significant outlier samples were identified in either the training set GSE75436 or the validation set GSE54083; therefore, all samples were retained for subsequent analyses.

The “limma” package was applied to screen for differentially expressed genes (DEGs) between the IA and control groups. Statistical significance was defined by an adjusted P-value (P.adj) < 0.05 coupled with an absolute log₂ fold change (|log₂FC|) ≥ 1. To visualize the distribution and expression patterns of DEGs, volcano plots and heatmaps were generated using the “ggplot2” and “pheatmap” R packages, respectively. Subsequently, the set of identified DEGs was intersected with the immune-related gene list retrieved from ImmPort to obtain immune-related differentially expressed genes (IM-DEGs) for downstream investigation.

Functional annotation and pathway enrichment analysis of IM-DEGs were performed using the “enrichGO” and “enrichKEGG” functions within the “clusterProfiler” R package. Gene Ontology (GO) analysis was conducted to categorize gene product functions across three domains: biological process (BP), molecular function (MF), and cellular component (CC) [8]. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was carried out to explore the involvement of IM-DEGs in pertinent signaling cascades and higher-order biological functions [9]. The most significantly enriched GO terms and KEGG pathways were prioritized and presented, with the significance threshold established at an adjusted P-value < 0.05.

To investigate potential interactions among the encoded proteins of IM-DEGs, the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (version 11.5) was utilized to construct a protein-protein interaction (PPI) network. A high-confidence interaction score threshold of ≥0.700 was applied, and disconnected nodes were concealed from the network view. The resulting network was subsequently visualized and analyzed using Cytoscape software (version 3.10.1). Hub genes were identified employing the CytoHubba plugin within Cytoscape, wherein the Maximal Clique Centrality (MCC) algorithm was applied to rank and select the top 10 pivotal genes [10].

The independent dataset GSE54083 served as the validation cohort for verifying the expression profiles of the identified hub genes. Violin plots illustrating expression distributions between control and IA groups were constructed using the “ggplot2” package. Statistical significance of intergroup expression differences was assessed according to the following procedure: First, the Shapiro-Wilk test was employed to evaluate the normality assumption of gene expression data distributions. For data conforming to a normal distribution, a two-tailed independent Student’s *t*-test was applied; for data deviating from normality, the Wilcoxon rank-sum test was utilized. A P-value < 0.05 is considered indicative of a significant difference. Hub genes demonstrating consistent and statistically significant differential expression in the validation set were retained for model construction. A linear predictive joint diagnostic model was developed using the “glmnet” package, with parameter optimization achieved through cross-validation. Receiver operating characteristic (ROC) curves were generated using the “pROC” package to evaluate diagnostic performance in both the training and validation cohorts. Performance metrics, including area under the curve (AUC), 95% confidence intervals (CI), sensitivity, and specificity, were computed.

Pearson correlation analysis was conducted among the core genes to assess their co-expression patterns using the “corrplot” R package. Correlation coefficients (Cor) were calculated, and the results were visualized via a correlation heatmap. The magnitude of the correlation coefficient reflects the strength of the association, with values closer to 1 indicating stronger correlations. Positive values denote a positive correlation, whereas negative values indicate an inverse relationship. Statistical significance for correlations was set at P < 0.01. This analysis aimed to elucidate potential synergistic regulatory relationships among core genes within the immune-inflammatory milieu of IA.

All data processing and statistical analyses were executed using R software (version 4.3.1) and associated bioinformatics packages. Quantitative variables are expressed as mean ± standard deviation. Prior to intergroup comparisons, the Shapiro-Wilk test was applied to assess data normality: the two-tailed independent Student’s *t*-test was used for normally distributed data, whereas the Wilcoxon rank-sum test was employed for non-normally distributed data. Significance thresholds were defined as follows: for DEG screening, |log₂FC| > 1 and adjusted P < 0.05; for functional enrichment, adjusted P < 0.05; and for gene correlation analysis, P < 0.01. Diagnostic model efficacy was assessed via ROC curve analysis, with AUC and corresponding 95% CI employed to quantify accuracy. All statistical tests were two-sided, and a P-value < 0.05 was uniformly considered statistically significant.

Following normalization of the training dataset GSE75436, boxplot visualization confirmed consistent expression distributions across all samples, indicating satisfactory inter-group comparability (Figure 1(a)). Principal component analysis (PCA) revealed a distinct separation between IA specimens and control tissues along the principal component axes, signifying substantial divergence in global gene expression profiles between the two groups (Figure 1(b)). Applying the predefined thresholds of |log₂ fold change (FC)| ≥ 1 and adjusted P-value < 0.05, a total of 757 DEGs were identified, comprising 389 up-regulated and 368 down-regulated transcripts. The overall distribution of DEGs and the expression patterns of the top 50 most significantly altered genes are depicted via a volcano plot and a heatmap, respectively (Figure 1(c) and Figure 1(d)). Concurrently, a compendium of 3118 immune-related genes was retrieved from the ImmPort database. Intersection of the 757 DEGs with this immune gene set yielded 173 IM-DEGs for subsequent investigation (Figure 1(e)).

GO enrichment analysis of the IM-DEGs was conducted using the “clusterProfiler” package in R to obtain standardized functional annotations across biological

(a)

(b) (c)

(d) (e)

Figure 1. Identification of differentially expressed genes. (a) Boxplot illustrating normalized expression distributions of the microarray dataset from IA patients. (b) Principal component analysis of the IA expression microarray dataset. (c) Volcano plot displaying the distribution of differentially expressed genes. (d) Heatmap depicting the expression profiles of the top 50 most significantly differentially expressed genes. (e) Venn diagram illustrating the overlap between DEGs and immune-related genes.

pathways, molecular functions, and cellular localizations. The top eight most significantly enriched terms within each GO category—biological process (BP), cellular component (CC), and molecular function (MF)—are presented in Figure 2(a) and Figure 2(b). The GO analysis indicated that IM-DEGs were predominantly enriched in immune-inflammatory biological processes, including leukocyte migration and immune response pathways. Regarding cellular localization, these genes were primarily associated with structures such as secretory granule membranes. In terms of molecular function, significant enrichment was observed for cytokine activity and chemokine activity. KEGG pathway analysis further demonstrated that these genes predominantly participate in immune-inflammatory cascades, notably cytokine-cytokine receptor interaction, chemokine signaling, and viral protein interaction with cytokine and cytokine receptor pathways (Figure 2(c) and Figure 2(d)). Collectively, these findings suggest that IM-DEGs contribute to the pathological progression of IA through modulation of immune-inflammatory responses, thereby establishing a foundation for subsequent hub gene prioritization and mechanistic exploration.

A protein-protein interaction (PPI) network encompassing the 173 IM-DEGs was constructed utilizing the STRING online resource. After concealing unconnected nodes, the resultant network comprised 156 nodes interconnected by 892 edges.

(a)

(b)

(c)

(d)

Figure 2. GO and KEGG enrichment analyses of IM-DEGs. (a) Bar plot illustrating the top enriched GO terms for IM-DEGs. (b) Bubble plot depicting GO enrichment results for IM-DEGs. (c) Bar plot displaying the top enriched KEGG pathways for IM-DEGs. (d) Bubble plot showing KEGG pathway enrichment results for IM-DEGs.

Visualization of the network was performed using Cytoscape software. Application of the Maximal Clique Centrality (MCC) algorithm via the CytoHubba plugin identified the top 10 highest-scoring hub genes, namely IL1B, TNF, CXCL8, CD8A, TYROBP, CCR1, CXCL10, CCL4, CCR5, and CCL20. Node coloration intensity corresponds to the MCC score, with darker hues denoting higher centrality. Among these, IL1B, TNF, CXCL8, CD8A, and TYROBP occupied the most central positions within the network, exhibiting the highest connectivity (Figure 3).

Expression levels of the five identified core genes (IL1B, TNF, CXCL8, CD8A, and TYROBP) were validated using the independent dataset GSE54083. Violin plots were generated to compare expression distributions between the IA and control groups. The findings are detailed below:

CXCL8: Expression in the IA group (test) was markedly elevated relative to the control group (ref), a difference that achieved statistical significance (P < 0.001). The violin plot (Figure 4(a)) illustrates a clear upward shift in the expression distribution of the IA cohort with minimal overlap between groups, implying that heightened CXCL8 expression may facilitate IA initiation and progression, potentially acting as a disease-promoting factor.

TNF: Conversely, TNF expression was significantly higher in the control group (ref) compared with the IA group (test) (P < 0.001). As depicted in Figure 4(b),

Figure 3. PPI network construction and hub gene identification. Visualization of the IM-DEG PPI network and identification of core hub genes.

the expression distribution of the control cohort is globally elevated relative to that of the aneurysm cohort, suggesting that diminished TNF expression is closely associated with IA pathogenesis and may function as a potential protective or inhibitory factor.

CD8A: Expression of CD8A in the control group (ref) significantly surpassed that in the IA group (test) (P < 0.05). The violin plot (Figure 4(c)) reveals a higher overall expression distribution in controls, indicating that downregulation of CD8A may reflect aberrant T-cell immune functionality within aneurysmal tissue, thereby contributing to immune-inflammatory dysregulation.

IL1B: Although expression levels in the control group (ref) tended to be higher than those in the IA group (test), the difference did not reach statistical significance (P > 0.05) (Figure 4(d)). This suggests that the expression discrepancy may be influenced by limitations in validation cohort sample size, platform-specific variations, or sample heterogeneity. The precise role of IL1B in IA pathogenesis warrants further investigation.

TYROBP: Expression in the IA group (test) appeared elevated relative to controls (ref); however, this trend was not statistically significant (P > 0.05) (Figure 4(e)). Given its strong correlation with other core genes such as CXCL8 and IL1B, TYROBP retains potential research value, and future studies with expanded sample sizes are needed to clarify its expression characteristics and functional relevance.

(a) (b)

(c) (d)

(e)

Figure 4. Expression analysis of core genes in the validation dataset. (a) Violin plot of CXCL8 expression in the validation cohort. (b) Violin plot of TNF expression in the validation cohort. (c) Violin plot of CD8A expression in the validation cohort. (d) Violin plot of IL1B expression in the validation cohort. (e) Violin plot of TYROBP expression in the validation cohort.

Pearson correlation analysis was performed among the five core genes (CXCL8, TNF, CD8A, TYROBP, and IL1B), with the results visualized as a correlation heatmap (Figure 5(a) and Figure 5(b)). The analysis revealed robust positive correlations among all core genes (P < 0.01), with correlation coefficients ranging from 0.46 to 0.78. The strongest association was observed between CXCL8 and TYROBP (Cor = 0.78), followed by IL1B and TNF (Cor = 0.71). Additional notable correlations included CD8A with TNF (Cor = 0.65), CXCL8 with IL1B (Cor = 0.58), and CD8A with TYROBP (Cor = 0.46). These significant positive interrelationships suggest that these core genes may operate synergistically within the immune-inflammatory regulatory network of IA, collectively contributing to disease onset and evolution.

(a) (b)

Figure 5. Correlation heatmap of core gene expression. (a) Heatmap displaying P-values for core gene correlation significance. (b) Heatmap depicting Pearson correlation coefficients among core genes.

To translate the identified core immune-related genes into a clinically applicable individualized diagnostic tool, a combined diagnostic model was developed based on the expression profiles of the core genes. A multi-dimensional validation framework was employed to systematically assess its diagnostic accuracy, robustness, and potential clinical utility. The results are presented below.

3.6.1. Evaluation of Single-Gene Diagnostic Performance

To ascertain the independent diagnostic utility of each core gene for IA, receiver operating characteristic ROC curves were generated using the training cohort GSE75436, and the area under the curve AUC was calculated to quantify discriminatory capacity. As illustrated in Figure 6(a), all five candidate core genes exhibited robust diagnostic performance for IA. The AUC values were as follows: TYROBP, 0.960 (95% CI: 0.885 - 1.000); IL1B, 0.933 (95% CI: 0.836 - 1.000); TNF, 0.884 (95% CI: 0.765 - 1.000); CD8A, 0.849 (95% CI: 0.715 - 0.983); and CXCL8, 0.813 (95% CI: 0.668 - 0.958). Each of these values surpassed the threshold of 0.8, indicative of excellent discriminatory power. These findings confirm that the core genes identified in this study represent potential biomarkers for IA diagnosis and provide a solid molecular foundation for constructing a multi-gene joint diagnostic model.

3.6.2. Construction of a Nomogram Model and Variable Selection

To identify variables with independent incremental diagnostic value for IA and to mitigate model overfitting, logistic regression was employed for variable selection. TYROBP was ultimately excluded due to a lack of independent diagnostic gain within the model, while the remaining four core genes—IL1B, CD8A, TNF, and CXCL8—were retained for nomogram construction (Figure 6(b)). The scoring of core genes corresponds to their clinically detectable expression ranges: IL1B corresponds to an expression interval of 6 - 12; CD8A, 2 - 12; TNF, 2.5 - 8.5; and CXCL8 corresponds to its full detection range. The total point scale ranges from 0 to 160, corresponding to a linear predictor value interval of −8 to 8. The total points map to an estimated probability of IA risk, ranging from 0.2 to 0.8.

It is important to clarify that the four-gene nomogram described above was derived from variable selection using logistic regression within the training set and is intended to illustrate the capacity of multi-gene combinations for quantitative disease risk assessment. However, to ensure model robustness and reproducibility in external independent validation, an additional cross-platform expression stability screening step was implemented: only core genes demonstrating statistically significant intergroup differential expression (P < 0.05) in the validation set GSE54083 were retained. Following this screening, IL1B was excluded because its expression difference in the validation set did not reach statistical significance; TYROBP, despite exhibiting the highest single-gene AUC, was also excluded for the same reason. Consequently, the primary final diagnostic model of this study is the three-gene combined model based on CXCL8, TNF, and CD8A. This model balances internal goodness-of-fit with external generalizability and constitutes the core conclusion of this investigation. The four-gene nomogram is presented as an internal exploratory analysis result but is not recommended as the final model.

3.6.3. Diagnostic Model Construction with Internal and External Validation

Integrating the differential expression results from the validation cohort, only CXCL8, TNF, and CD8A demonstrated statistically significant inter-group differences in GSE54083, thereby exhibiting robust cross-cohort expression stability. Consequently, these three genes were selected as core variables to construct a precise and reproducible joint diagnostic model, ensuring external reliability and potential for clinical translation. ROC curve analyses were performed in both the training set GSE75436 and the independent validation set GSE54083.

Internal validation (Training set GSE75436): The model achieved an AUC of 0.947 (95% CI: 0.862 - 1.000), demonstrating exceptional discriminatory capacity to effectively distinguish IA samples from normal controls. This indicates a high degree of model fit and diagnostic reliability within the training cohort (Figure 6(c)).

External validation (Independent validation set GSE54083): The model yielded an AUC of 0.754 (95% CI: 0.546 - 0.962). Although the external validation performance was moderately attenuated relative to internal validation—likely attributable to cross-platform technical variation and sample size constraints of the validation cohort—the model maintained a stable level of discriminatory ability. This corroborates the favorable cross-cohort robustness and generalizability of the diagnostic model constructed from these three core genes (Figure 6(d)).

(a)

(b)

(c) (d)

Figure 6. Diagnostic model construction with internal and external validation. (a) Single-gene ROC curve analysis for the five core genes. (b) Nomogram depicting individualized risk estimation for intracranial aneurysm. (c) ROC curve analysis of the three-gene combined diagnostic model in the GSE75436 training set. (d) External validation ROC curve of the three-gene combined diagnostic model in the independent GSE54083 dataset.