Table 1.
National center for biotechnology information gene expression omnibus datasets information.
Fig 1.
Identification of differentially expressed genes in ischemic stroke and depression compared with normal controls.
(a) x- and y-axes denote logFC gene expression fold change and -log₁₀(p-value) statistical significance, respectively. Negative and positive logFC indicate downregulation and upregulation, respectively. (b) Heatmap of the top 100 differentially expressed genes. (c) x- and y-axes denote logFC gene expression fold change and -log₁₀(p-value) statistical significance, respectively. Negative and positive logFC indicate downregulation and upregulation, respectively. (d) Heatmap of the top 100 differentially expressed genes.
Fig 2.
Identifying key hub genes through target mapping, functional enrichment, and network analysis.
(a) Potential ketamine targets obtained from the CTD, GeneCards, and Swiss Target Prediction database. (b) Venn diagram showing the intersections among DEGs1, DEGs2, and putative ketamine targets. (c) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment results for intersection genes. Sankey diagram shows gene–pathway relationships. Dot plot x- and y-axes represent the number of enriched genes and enriched KEGG pathways, respectively. Dot size and color indicates gene count and p-value magnitude, respectively. (d) GO enrichment results for intersection genes. The top 8 results per category are shown. (e) Protein interaction network of intersection genes. Node size and color indicate Degree: larger Degree values signify greater importance within the network. (f) Venn diagram of feature gene intersections identified by the MCC, Degree, EPC, and MNC algorithms.
Fig 3.
SHAP-based interpretability analysis and feature ranking for the ischemic stroke model (GSE16561).
(a) Bar plot of the SHAP-based feature importance ranking in the ischemic stroke model, with genes ordered by descending mean absolute SHAP value. (b) Beeswarm plot illustrating the distribution and impact direction of SHAP values for the top 14 features in the ischemic stroke model, ordered vertically by descending mean absolute SHAP value. (c) Waterfall plot decomposing an individual ischemic stroke prediction into additive SHAP contributions from the top features, starting from the baseline E[f(x)] = 0.619. (d) Force plot illustrating how feature contributions collectively shift an individual ischemic stroke prediction from the baseline E[f(x)] = 0.619 to the final output f(x) = 0.981.
Fig 4.
SHAP-based interpretability analysis and feature ranking for the depression model (GSE23848).
(a) Bar plot of the SHAP-based feature importance ranking in the depression model, with genes ordered by descending mean absolute SHAP value. (b) Beeswarm plot showing the distribution and direction of SHAP values for the top 14 features in the depression model, ordered vertically by descending mean absolute SHAP value. (c) Waterfall plot decomposing an individual depression model prediction into additive SHAP contributions, starting from the baseline E[f(x)] = 0.571. (d) Force plot illustrating how feature contributions collectively shift an individual depression model prediction from the baseline E[f(x)] = 0.571 to the final output f(x) = 0.997.
Fig 5.
Expression of hub genes in ischemic stroke training(a) GSE16561 and validation (b) GSE58294 sets.
X- and y-axes represent genes and expression levels, respectively. p-values were estimated by the Wilcoxon rank-sum test: *p < 0.05, ***p < 0.001, ****p < 0.0001). Expression of hub genes in depression training (c) (GSE23848) and validation (d) (GSE76826) sets. X- and y-axes represent genes and expression levels, respectively. p-values were estimated by the Wilcoxon rank-sum test: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). Receiver operating characteristic (ROC) curves of hub genes in ischemic stroke training (e) GSE16561 and validation (f) GSE58294 sets. X- and y-axes represent false and true positive rates, respectively. ROC curves of feature genes in depression training (g) GSE23848 and validation (h) GSE76826 sets. X- and y-axes represent false and true positive rates, respectively.
Fig 6.
Correlation among biomarkers in (a) ischemic stroke training and (b) validation sets.
(c) and (d) show depression training and validation sets, respectively. Correlation coefficients and p-values were derived from the Pearson correlation test.
Fig 7.
Gene set enrichment analysis (GSEA) of DDIT3 and IL1RN reveals significant pathway associations in ischemic stroke and depression.
(a) GSEA results for DDIT3 in ischemic stroke. The two highest and lowest enriched pathways are shown. Line plot shows enrichment score (ES) for ranked gene sets, barcode plot marks positions of pathway gene set members in the ranked gene list, and heatmap shows the distribution of rank values for all genes post-ranking. (b) GSEA enrichment results for IL1RN in ischemic stroke. The two highest and lowest enriched pathways are shown. Line plot shows ES, barcode plot marks gene positions, and heatmap shows rank value distribution. (c) GSEA enrichment results for DDIT3 in depression. The two highest and lowest enriched pathways are shown. Line plot shows ES, barcode plot marks gene positions, and heatmap shows rank value distribution. (d) GSEA enrichment results for IL1RN in depression. The two highest and lowest enriched pathways are shown. Line plot shows ES, barcode plot marks gene positions, and heatmap shows rank value distribution.
Fig 8.
Comprehensive landscape of immune infiltration and its correlation with key biomarkers in ischemic stroke and depression.
(a) Immune infiltration in the ischemic stroke training set assessed by CIBERSORT. X- and y-axes represent different samples and infiltration proportion, respectively. (b) Immune infiltration in the depression training set assessed by CIBERSORT. X- and y-axes represent different samples and infiltration proportion, respectively. (c) Differences in immune infiltration abundance between ischemic stroke disease and normal control samples. p-values estimated by the t-test: *p < 0.05, **p < 0.01, ****p < 0.0001. (d) Differences in immune infiltration abundance between depression disease and normal control samples. p-values estimated by the t-test: *p < 0.05, **p < 0.01, ****p < 0.0001. (e) Correlation between biomarkers and differentially infiltrated immune cells in ischemic stroke. X- and y-axes list immune cells and genes, respectively. Correlation coefficients and p-values estimated by the Spearman correlation test: *p < 0.05, **p < 0.01, ***p < 0.001. (f) Correlation between biomarkers and differentially infiltrated immune cells in depression. X- and y-axes list immune cells and genes, respectively. Correlation coefficients and p-values estimated by the Spearman correlation test: *p < 0.05, **p < 0.01, ***p < 0.001.
Fig 9.
SIGNOR network of biomarkers with confidence scores ≥ 0.7.
Red and blue indicate inhibitory and activating pathways, respectively.
Table 2.
Molecular docking results.
Fig 10.
3D chemical structures of (a) DDIT3 and (b) IL1RN biomarkers. (c) 3D chemical structure of ketamine. (d) Molecular docking results for DDIT3 and ketamine. (e) Molecular docking results for IL1RN and ketamine.