Identification immune-related hub genes in diagnosing atherosclerosis with ischemic stroke through comprehensive bioinformatics analysis and machine learning

Abstract

Background: Atheroma plaques are major etiological factors in the pathogenesis of ischemic stroke (IS). Emerging evidence highlights the critical involvement of the immune microenvironment and dysregulated inflammatory responses throughout IS progression. Consequently, therapeutic strategies targeting specific immune-related markers or signaling pathways within this microenvironment hold significant promise for IS management.

Methods: We integrated Weighted Gene Co-expression Network Analysis (WGCNA), CIBERSORT, and machine learning (LASSO/Random Forest) to identify disease-associated modules and hub genes. Immune infiltration analysis evaluated hub gene-immune cell correlations, while protein-protein interaction (PPI) and ROC curve analyses assessed diagnostic performance.

Results: Comprehensive bioinformatics analysis identified three hub genes-OAS2, TMEM106A, and ABCB1-with high prognostic value for ischemic stroke. Immune infiltration profiling revealed significant correlations between these genes and distinct immune cell populations, underscoring their roles in modulating the immune microenvironment. The diagnostic performance of the gene panel was robust, achieving an area under the curve (AUC) was calculated as 0.9404 (p < 0.0001; 95% CI: 0.887-0.9939) for atherosclerotic plaques, demonstrating superior accuracy compared to conventional biomarkers.

Conclusion: By integrating machine learning with multi-omics bioinformatics, we established a novel three-gene signature (OAS2, TMEM106A, ABCB1) for precise diagnosis of atherosclerosis and ischemic stroke. These genes exhibit dual diagnostic utility and may influence disease progression through immune cell modulation. Our findings provide a foundation for developing targeted therapies and biomarker-driven clinical tools.

Keywords: WGCNA; atherosclerosis; bioinformatics; immune cell infiltration; ischemic stroke; machine learning.

Figure 1

Overview of the study design.

Figure 2

Functional enrichment analysis and the immune infiltration landscape of differential expression gene. (A) the volcano of differentially expressed genes between Atheroma plaque compared with normal tissue. A p < 0.05 and |log FC (fold change)| > 0.25 were considered statistically significant. Grenn blot represent downregulated and pink blot represent upregulated. (B) To analyze potential mRNA targets, using the Gene Ontology (GO) system. The ClusterProfiler utility within R software facilitated the clustering of prospective targets based on biological pathways (BP), molecular functions (MF), and cellular components (CC). A significance threshold of q value <0.05 was applied to determine statistical significance in the enrichment results. (C) Enriched KEGG signaling pathways were selected to illustrate the significant biological activities associated with potential mRNA. The gene ratio is represented on the abscissa, while the enriched pathways are depicted on the ordinate. (D, E) The proportion of 22 immune cell subpopulations in 64 samples from the GSE43292 datasets. (F) The disparity in immune infiltration between atheroma plaque and healthy controls was examined. The normal controls group was color-coded red, while the atheroma plaque group was color-coded green. Statistical significance was defined as a P < 0.05.

Figure 3

Construction of the co-expression network. (A) Soft threshold (power = 20) and scale-free topology fit index (R² = 0.9). (B) The correlation heatmap of different modules. (C) Gene hierarchy tree clustering diagram. The graph indicates different genes horizontally and the uncorrelatedness between genes vertically, the lower the branch, the less uncorrelated the genes within the branch, i.e., the stronger the correlation. (D) Heatmap showing the relations between the modules and Ap feature. The value in the small cells of the graph represent the two-calculated correlation values cor coefficients between the eigenvalues of each trait and each module as well as the corresponding statistically significant p-values. Color corresponds to the size of the correlation; the darker the red, the more positive the correlation; the darker the green, the more negative the correlation. (E) Venn diagram for identification the overlapping genes from DEGs and WGCNA modules.

Figure 4

The final hub genes were identified by lasso regression and random forest analyses. (A, B) A least operator shrinkage and selection operator (LASSO) logistic regression was used to screen characteristic variables. (C) Ordination plot of gene importance scores. (D) Venn diagram showing the intersection feature variables filtered by the two algorithms. (E) The correlation between hub genes and the immune status was obtained by CIBERSORT algorithm analysis.

Figure 5

Protein-protein interaction network and diagnostic efficacy of OAS2. (A–C) protein-protein interaction network of OAS2, ABCB1, TMEM106A; (D) Receiver operating characteristic (ROC) curves assessing the diagnostic efficacy of OAS2, TMEM106A and ABCB1 in GSE43292 dataset. (E) Receiver operating characteristic (ROC) curves assessing the diagnostic efficacy of OAS2, TMEM106A and ABCB1 in GSE22255 dataset.

Figure 6

Nomogram for predicting the risk of ischemic stroke based on key gene expression.