Fig 1.
The flowchart of this study.
Table 1.
Basic information of the GSE113439, GSE117261, and GSE53408 datasets.
Table 2.
qRT-PCR primer sequence.
Fig 2.
Assessment of batch effects and harmonization of integrated PAH transcriptomic datasets.
(A) Boxplots of normalized gene-expression values across individual samples before batch-effect correction, colored by study origin (GSE113439, blue; GSE117261, red); (B) UMAP projection of uncorrected expression data, demonstrating dataset-driven separation of samples, indicating batch effects. Each dot represents one sample; (C) Boxplots of gene-expression distributions after batch-effect correction; (D) UMAP projection after batch-effect correction, showing improved inter-dataset integration and reduced dataset-driven clustering.
Fig 3.
Immune-cell infiltration analysis in PAH and control lung tissues.
(A) Heatmap illustrating the relative abundance of 24 immune-cell subsets across individual samples, as inferred by ssGSEA; (B) Violin plots showing differential comparison of immune-cell infiltration between PAH and control lung tissues using the Wilcoxon rank-sum test; (C) LASSO regression coefficient profiles of immune-cell subsets across varying penalty parameters (λ); (D) LASSO coefficient profiles of immune-cell subsets across varying values of λ, illustrating shrinkage and selection of immune-cell features associated with PAH classification.
Fig 4.
WGCNA of integrated PAH transcriptomic data.
(A) Sample clustering dendrogram of the merged dataset with corresponding clinical trait annotation (Control vs. PAH), used to detect potential outliers prior to network construction; (B) Determination of the optimal soft-thresholding power for network construction; (C) Gene dendrogram and co-expression modules identified by dynamic tree cut (colored bars); (D) Heatmap of module-trait relationships, showing correlations between module eigengenes and clinical phenotype (Control vs. PAH). Correlation coefficients and corresponding p-values are displayed in each cell.
Fig 5.
Identification of DEGs and GO enrichment analysis in PAH lung tissues.
(A) Volcano plot of DEGs between PAH and control samples. Red: upregulated; blue: downregulated; and gray: non-significant; x-axis: log2 fold change; and y-axis: -log10 (adjusted p value); (B) Heatmap of DEGs across individual samples, illustrating distinct expression patterns between PAH and control groups; (C) Bubble plot of GO enrichment results (BP, CC, and MF categories); x-axis: gene ratio; bubble size: gene counts; (D-F) Circular network plots illustrating the relationships between enriched GO terms and associated genes for BP (D), CC (E), and MF (F), respectively.
Fig 6.
PPI network construction and module identification of DEGs.
(A) Global PPI network of DEGs using the STRING database. Nodes: proteins; edges: predicted or experimentally validated interactions; (B) Visualization of the PPI network in Cytoscape software, highlighting hub genes and network topology; (C-E) Highly interconnected gene clusters identified using the MCODE algorithm, representing core functional modules within the PPI network.
Fig 7.
Hub gene identification and the co-expression network involving mRNA and target miRNA.
(A) Workflow for screening hub genes; (B) PPI network (constructed using the STRING software) of the eight identified hub genes; (C) The mRNA-miRNA co-expression network.
Fig 8.
External validation of hub gene expression in the GSE53408 dataset.
(A-H) Violin plots showing the expression levels of eight hub genes in PAH and control lung tissues from the independent validation cohort (GSE53408). Statistical significance between groups was assessed using the Wilcoxon rank-sum test; (I) Heatmap of hub gene expression, demonstrating distinct clustering between PAH and control groups.
Fig 9.
Diagnostic performance of hub genes in PAH lung tissues.
(A) The diagnostic value of 8 hub genes in PAH was evaluated by the GSE53408 datasets; (B) Ten-fold cross-validation results; (C) The diagnostic value of KDM6A and CTNNB1 in PAH; (D) The diagnostic value of CHD8 and RBM391 in PAH; (E) The diagnostic value of CDC5L and ASH1L in PAH; (F) The diagnostic value of SMARCA5 and BCLAF1 in PAH.
Fig 10.
Machine learning-based prioritization and SHAP interpretability analysis of hub genes.
(A) Heatmap comparing model performance across different machine learning algorithms, showing AUC values in the training cohort and external validation cohort (GSE53408); (B) Feature importance ranking based on mean absolute SHAP values, indicating the relative contribution of each hub gene to model prediction; (C) SHAP summary (beeswarm) plot illustrating the distribution of SHAP values for each gene across samples. Each point represents one sample, and color indicates feature expression level (high to low); (D) SHAP waterfall plot for a representative sample, demonstrating the individual contribution of each gene to the final prediction probability.
Fig 11.
Correlation analysis between hub gene expression and immune-cell signatures in PAH lung tissues.
(A) Heatmap of Spearman correlation coefficients between hub genes and inferred immune-cell subsets. Color intensity represents the strength and direction of correlation; (B-C) Positive correlations between CDC5L and RBM39 expression levels and Tgd signatures; (D) Negative correlation between ASH1L expression and neutrophil signatures; (E-F) Negative correlations between CTNNB1 expression and aDC and pDC; (G) Negative correlation between SMARCA5 expression and Tcm signatures. Spearman correlation coefficients (R) and corresponding p values are indicated in each panel.
Fig 12.
Hemodynamic validation of the PAH model.
(A) Right ventricular systolic pressure (RVSP); (B) Mean pulmonary artery pressure (mPAP); (C) Right ventricular hypertrophy index (RVHI) in control and model mice. Data are presented as mean ± SD. Statistical comparisons between groups were performed using an unpaired Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, no significant difference.
Fig 13.
Comparison of organ weights between control and PAH model mice.
(A-F) Absolute organ weights of lung (A), heart (B), liver (C), spleen (D), kidney (E), and brain (F) in control and model groups. Data are presented as mean ± SD. Statistical comparisons between groups were performed using an unpaired Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, no significant difference.
Fig 14.
Histological assessment of lung tissues in control and PAH model mice.
(A) Representative HE staining and Masson’s trichrome staining images of lung sections from control and PAH model groups. Scale bar = 100 μm; (B) Quantification of collagen area fraction based on Masson staining. Data are presented as mean ± SD. Statistical comparison was performed using an unpaired Student’s t-test. ***p < 0.001.
Fig 15.
Validation of hub gene expression in lung tissues from control and PAH model mice by qRT-PCR.
A-H. Relative mRNA expression levels of the eight hub genes (RBM39, BCLAF1, CDC5L, CTNNB1, SMARCA5, CHD8, ASH1L, and KDM6A) were measured in lung tissues from control and SuHx-induced PAH mice using qRT-PCR. Data are presented as mean ± SD. Statistical comparisons between groups were performed using an unpaired Student’s t-test. ns, not significant; ***p < 0.001; ****p < 0.0001.