Early diagnosis of acute myocardial infarction via hub genes identified by integrated weighted gene co-expression network analysis

Fig. 1. Flow chart of the study process. WGCNA, weighted gene co-expression network analysis; GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; LASSO, least absolute shrinkage and selection operator; SVM-RFE, support vector machine-recursive feature elimination; RF, random forest; GSEA, gene set enrichment analysis.

Fig. 2. Data processing and normalization of the GSE66360 dataset. The principal component analysis (PCA) plot before and after outliers were removed and robust multiarray averaging(RMA) normalization in the discovery (A, B) and validation (C, D) cohorts of the GSE66360 dataset. A box plot shows the gene expression of each sample before and after adjustment of Between Arrays function of linear models normalized in the discovery (E, F) and validation (G, H) cohorts of the GSE66360 dataset.

Fig. 3. Identification of CEC-derived AMI-associated genes through DEGs and WGCNA. (A) Volcano plots display the differentially expressed genes (DEGs) of circulating endothelial cells (CECs) between acute myocardial infarction (AMI) patients and health population. The red nodes represent upregulated genes, and the turquoise nodes represent downregulated genes. (B, C) Analyzed network topology for various soft thresholding powers to construct a scale-free co-expression network. (D) The hierarchical dendrogram cut at a height of 128 for outlier identification. (E) Merge network modules with a minimum module size at 50 and a cut tree height at 0.5. (F) The heatmap shows the correlations between each module and AMI patients or healthy controls. (G, H) A scatterplot of Gene Significance (GS) for AMI vs. Module Membership (MM) in both modules. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4. Functional enrichment analysis of intersection genes and construction of the PPI network. (A) Venn diagram indicates the distribution of the weighted gene co-expression network genes from pink and turquoise modules and differentially expressed genes (DEGs). (B, C) Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses of intersection genes. (D) The protein-protein interaction (PPI) network based on intersection genes was constructed, and the hub gene cluster with the highest scores including 37 nodes were screened. The nodes represent genes, and the edges represent links between the genes. (E, F) GO and KEGG analyses of the identified hub gene cluster. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5. Selection of hub characteristic genes via three machine learning methods. (A) The least absolute shrinkage and selection operator (LASSO) regression model was constructed, and the eight most powerful predicted genes were selected. The red dots and gray lines represent the binomial deviance value and standard error, respectively, while the vertical dotted lines represent optimal values with the minimum criteria and 1-SE criteria. (B) The plot displays coefficients of 1-SE criteria for each gene in the LASSO regression model. (C) The results of the support vector machine-recursive feature elimination (SVM-RFE) algorithm for feature selection. The X-axis represents the number of feature genes, and the Y-axis represents the accuracy of the model. (D) The random forest (RF) classification model was constructed, and the plot showed the relationships between the number of trees and accuracy rate. The green, red, and blue folding lines represent the variations in mis-classification rates with respect to the number of decision trees for all individuals, patients with MI, and healthy controls in the discovery cohort, respectively. (E) Venn diagram presents the shared hub genes determined…

Fig. 6. Diagnostic efficacy of hub characteristic genes in the prediction of acute myocardial infarction and gene verification. (A, B) ROC curves of the diagnostic performance of hub characteristic genes in the discovery and validation cohorts. (C, D) Box plot displays the mRNA expression level of hub characteristic genes in AMI patients and healthy controls in the discovery and validation cohorts.

Fig. 7. GSEA for signaling pathways involved in hub characteristic genes. (A–F) Enrichment plots of top 5 signaling pathways for hub characteristic genes including (A) CCL20, (B) CXCL16, (C) IL1R2, (D) LILRA1, (E) TYROBP, and (F) NFKBIA. LILRA1, leukocyte immunoglobulin-like receptor A1; CCL20, C—C motif chemokine ligand 20; IL1R2, interleukin 1 receptor type 2; TYROBP, transmembrane immune signaling adaptor; CXCL16, C-X-C motif chemokine ligand 16; NFKBIA, nuclear factor kappa-B inhibitor alpha.