An Integrative Transcriptomic, Network Pharmacology, and Molecular Docking Analysis of the Ferroptosis–Fibrosis Axis in Cardiomyopathy with Exploratory Relevance to Diabetic Cardiomyopathy

Figures

• 

  Figure 1. Transcriptomic profiling and differential expression analysis in GSE5406. (A) Volcano plot illustrating differentially expressed genes between nonfailing and cardiomyopathy samples. Red and blue points denote significantly upregulated and downregulated genes, respectively (adjusted p-value < 0.05). (B) Mean–difference (MA) plot showing the relationship between log2 fold-change and average expression levels. (C) UMAP plot demonstrating sample distribution and partial separation between control and disease groups. (D) Density plot showing comparable global expression distributions across samples, consistent with appropriate normalization. (E) Mean–variance trend plot derived from limma analysis, illustrating variance stabilization across expression levels; the blue horizontal line indicates the overall mean residual standard deviation estimated by the limma package. (F) Distribution of adjusted p-values across all probes, reflecting multiple testing correction and overall statistical behavior.

• 

  Figure 2. STRING-based protein–protein interaction network and functional enrichment analysis of the differentially expressed gene set. (A) Protein–protein interaction (PPI) network generated using the STRING database for the DEG-derived input list. Nodes represent proteins, and edges represent known or predicted functional associations. (B) Gene Ontology Biological Process (GO-BP) enrichment analysis. (C) Gene Ontology Cellular Component (GO-CC) enrichment analysis. (D) Gene Ontology Molecular Function (GO-MF) enrichment analysis. (E) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. (F) Reactome pathway enrichment analysis. In the enrichment plots, bubble size represents the number of genes associated with each term, and color intensity represents false discovery rate (FDR)-adjusted significance.

• 

  Figure 3. PPI subnetwork of hub genes identified using the MCC algorithm in the cytoHubba plugin of Cytoscape. Node color represents MCC scores, with red corresponding to higher centrality and yellow to lower centrality.

• 

  Figure 4. Disease enrichment analysis of differentially expressed genes using DisGeNET and OMIM databases. (A) Bar plot showing the top enriched disease terms identified from the DisGeNET database, ranked according to adjusted p-values. The analysis highlights associations with cardiovascular conditions (including congestive heart failure and myocardial infarction), connective tissue disorders (such as Ehlers–Danlos syndrome), and hemoglobin-related traits. Darker colors indicate higher statistical significance (higher −log10 adjusted p-values), whereas lighter colors indicate lower statistical significance. (B) Clustergram illustrating the relationship between selected enriched disease terms and input genes. Hemoglobin-related genes (HBB, HBA1, HBA2) cluster with erythroid and thalassemia-associated terms, while ECM–related genes (COL1A1, COL1A2, COL3A1, LUM) show associations with connective tissue–related conditions. These patterns suggest that the identified gene set may be linked to both erythroid and ECM–associated disease processes.

• 

  Figure 5. Pathway enrichment analysis based on KEGG 2026 and Reactome 2024 databases. (A) KEGG pathway enrichment results showing the top significantly enriched pathways, including cytoskeleton organization in muscle cells, ECM–receptor interaction, AGE–RAGE signaling pathway in diabetic complications, and the KEGG pathway Proteoglycans in cancer. (B) Reactome pathway enrichment analysis demonstrating enrichment of ECM-related processes, including ECM proteoglycans, ECM organization, and collagen-related pathways. Across both databases, enrichment patterns consistently point toward processes involving ECM organization, proteoglycan-associated pathways, and cytoskeletal structure, accompanied by signals related to inflammatory and metabolic processes. These findings provide a systems-level context for the observed gene expression changes. Darker colors indicate higher statistical significance (higher −log10 adjusted p-values), whereas lighter colors indicate lower statistical significance.

• 

  Figure 6. 3D docking conformations of candidate compounds within the predicted binding regions of TGFBR1, STAT3, GPX4, and AKT1. (A) Finerenone–TGFBR1, (B) Pirfenidone–TGFBR1, (C) Ruxolitinib–TGFBR1, (D) Deferiprone–TGFBR1, (E) Liproxstatin-1–TGFBR1, (F) Bardoxolone methyl–TGFBR1, (G) Finerenone–STAT3, (H) Pirfenidone–STAT3, (I) Ruxolitinib–STAT3, (J) Deferiprone–STAT3, (K) Liproxstatin-1–STAT3, (L) Bardoxolone methyl–STAT3, (M) Finerenone–GPX4, (N) Pirfenidone–GPX4, (O) Ruxolitinib–GPX4, (P) Deferiprone–GPX4, (Q) Liproxstatin-1–GPX4, (R) Bardoxolone methyl–GPX4, (S) Finerenone–AKT1, (T) Pirfenidone–AKT1, (U) Ruxolitinib–AKT1, (V) Deferiprone–AKT1, (W) Liproxstatin-1–AKT1, and (X) Bardoxolone methyl–AKT1. Representative low-energy docking poses are shown within the predicted binding regions of the respective target proteins. Ligands are displayed in stick representation, whereas surrounding protein structures are visualized using backbone, residue-level, and semi-transparent molecular surface representations to facilitate inspection of ligand orientation and local binding-site geometry. Docking conformations were generated using AutoDock 4.2.6 and visualized using PyMOL.

• 

  Figure 7. 3D docking conformations of candidate compounds within the predicted binding regions of SMAD3 and ACSL4. (A) Finerenone–SMAD3, (B) Pirfenidone–SMAD3, (C) Ruxolitinib–SMAD3, (D) Deferiprone–SMAD3, (E) Liproxstatin-1–SMAD3, (F) Bardoxolone methyl–SMAD3, (G) Finerenone–ACSL4, (H) Pirfenidone–ACSL4, (I) Ruxolitinib–ACSL4, (J) Deferiprone–ACSL4, (K) Liproxstatin-1–ACSL4, and (L) Bardoxolone methyl–ACSL4. Representative low-energy docking poses are shown within the predicted binding regions of the respective target proteins. Ligands are displayed in stick representation, whereas surrounding protein structures are visualized using backbone, residue-level, and semi-transparent molecular surface representations to facilitate inspection of ligand orientation and local binding-site geometry. Docking conformations were generated using AutoDock 4.2.6 and visualized using PyMOL.

• 

  Figure 8. Receiver operating characteristic (ROC) curves of ten candidate hub genes in the GSE5406 dataset. ROC curves were generated using normalized gene expression values to evaluate the discriminatory performance of candidate genes between cardiomyopathy and nonfailing myocardial samples. LUM and ASPN showed the highest AUC values among the evaluated genes. The dashed diagonal line represents the line of no discrimination (AUC = 0.5), corresponding to the performance expected by random classification.

• 

  Figure 9. Venn diagram showing the overlap of candidate genes identified by least absolute shrinkage and selection operator (LASSO) regression, random forest classification, and support vector machine–recursive feature elimination (SVM-RFE) in the GSE5406 dataset. Four genes (FCN3, HOPX, CNN1, and GLUL) were consistently selected by all three machine learning algorithms and were therefore defined as the core machine learning signature. Additional genes were shared between two algorithms, further supporting convergent machine-learning prioritization of candidate cardiomyopathy-associated features. Feature-selection results should be interpreted as exploratory and hypothesis-generating. Numbers within each region indicate the number of genes uniquely or jointly identified by the corresponding machine learning methods.

• 

  Figure 10. Exploratory evaluation analyses of the machine-learning–derived gene signature. (A) Internal validation of the machine-learning framework using stratified class-weighted five-fold cross-validation. To address potential overfitting and class imbalance within the discovery cohort, internal validation was performed using out-of-fold evaluation. The model achieved a cross-validated AUC of 0.934 (95% CI: 0.820–1.000). Class weighting was applied to account for the imbalance between cardiomyopathy samples (n = 194) and nonfailing control samples (n = 16). (B) Cross-cohort comparison of differential-expression patterns between the GSE5406 discovery cohort and the independent GSE263297 cohort. Each point represents a gene identified in the GSE5406 discovery analysis and present in the GSE263297 dataset. The x-axis indicates logFC values in GSE5406, whereas the y-axis indicates logFC values in GSE263297. Dashed lines denote zero logFC. Green points indicate genes with consistent direction of expression change across cohorts, whereas red points indicate discordant genes. Among the 51 genes evaluated in both cohorts, 20 exhibited the same direction of expression change, while one gene…