Immune-associated pivotal biomarkers identification and competing endogenous RNA network construction in post-operative atrial fibrillation by comprehensive bioinformatics and machine learning strategies

Abstract

Background: Atrial fibrillation (AF) is the most common arrhythmia. Previous studies mainly focused on identifying potential diagnostic biomarkers and treatment strategies for AF, while few studies concentrated on post-operative AF (POAF), particularly using bioinformatics analysis and machine learning algorithms. Therefore, our study aimed to identify immune-associated genes and provide the competing endogenous RNA (ceRNA) network for POAF.

Methods: Three GSE datasets were downloaded from the GEO database, and we used a variety of bioinformatics strategies and machine learning algorithms to discover candidate hub genes. These techniques included identifying differentially expressed genes (DEGs) and circRNAs (DECs), building protein-protein interaction networks, selecting common genes, and filtering candidate hub genes via three machine learning algorithms. To assess the diagnostic value, we then created the nomogram and receiver operating curve (ROC). MiRNAs targeting DEGs and DECs were predicted using five tools and the competing endogenous RNA (ceRNA) network was built. Moreover, we performed the immune cell infiltration analysis to better elucidate the regulation of immune cells in POAF.

Results: We identified 234 DEGs (82 up-regulated and 152 down-regulated) of POAF via Limma, 75 node genes were visualized via PPI network, which were mainly enriched in immune regulation. 15 common genes were selected using three CytoHubba algorithms. Following machine learning selection, the nomogram was created based on the four candidate hub genes. The area under curve (AUC) of the nomogram and individual gene were all over 0.75, showing the ideal diagnostic value. The dysregulation of macrophages may be critical in POAF pathogenesis. A novel circ_0007738 was discovered in POAF and the ceRNA network was eventually built.

Conclusion: We identified four immune-associated candidate hub genes (C1QA, C1R, MET, and SDC4) for POAF diagnosis through the creation of a nomogram and evaluation of its diagnostic value. The modulation of macrophages and the ceRNA network may represent further therapy methods.

Keywords: bioinformatics analysis; competing endogenous RNA network; diagnosis; immune and inflammation; machine learning; post-operative atrial fibrillation.

Figure 1

Study flowchart. POAF, post-operative atrial fibrillation; DEGs, differentially expressed genes; Limma, linear models for microarray data; Lasso, least absolute shrinkage and selection operator; SVM-RFE, support vector machine recursive feature elimination; DECs, differentially expressed circRNAs; CCRD, cancer-specific circRNA database; ceRNA, competing endogenous RNA.

Figure 2

The heatmap and volcano plot of DEGs for POAF compared with SR in GSE143924. (A) Heatmap reveals top 25 up- and down-regulated DEGs for POAF patients compared with SR. Red and black represent up- and down- expression, respectively. (B) The volcano plot shows all DEGs for POAF compared with SR. Red and green triangles represent the significant DEGs following the filtration criteria. SR, sinus rhythm; others see Figure 1 .

Figure 3

PPI network construction and common genes selection. (A) 75 POAF node genes were visualized from the PPI network. (B-D) Top 20 node genes were visualized using “Betweenness”, “Closeness”, and “Degree” algorithms via CytoHubba plug-in from Cytoscape, respectively. (E) The Venn diagram of top 20 node genes from three algorithms shows that 15 genes were identified as common genes for machine learning analysis after intersection. PPI, protein-protein interaction network; others see Figure 1 .

Figure 4

Functional enrichment analysis of POAF node genes. (A-C) Partial visualization of GO analysis for node genes from biological process, cellular component, and molecular function, respectively. X-axis represents gene ratio, Y-axis represents different ontologies, the circle color represents P-value and the circle size shows count number. (D) KEGG pathway analysis of node genes. The left side displays genes, while the right side depicts significant enriched pathways. The connection between genes and pathways refers to that genes were enriched in related pathway. POAF, post-operative atrial fibrillation; GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Figure 5

Machine learning algorithms for identifying candidate hub genes in diagnosing POAF. (A) Common genes were ranked based on the average rank using SVM-RFE algorithm after 100 folds. The lower the average rank, the greater the significance of the gene. (B) The number of genes (n=7) corresponding to the lowest point of the curve is the most suitable number for POAF diagnosis using Lasso algorithm. (C) The column reveals that genes were ranked with importance score using random forest algorithm. (D) The Venn diagram depicts the intersection of genes shared by three distinct methods. Five genes were chosen for nomogram development and diagnostic value assessment. POAF, post-operative atrial fibrillation; SVM-RFE, support vector machine-recursive feature elimination.

Figure 6

Nomogram construction and ROC evaluation. (A) The ROC curve of the individual candidate hub gene (C1QA, C1R, MET, SDC4) in nomogram construction. (B) The ROC curve of the same candidate hub gene derived from a different test dataset-GSE62871. (C) The nomogram was constructed based on the four candidate hub genes. (D) The ROC curve of the nomogram. ROC, receiver operating curve; C1QA, Complement C1qA Chain; C1R, Complement C1r; MET, MET Proto-Oncogene, Receptor Tyrosine Kinase; SDC4, Syndecan 4.

Figure 7

The heatmap and volcano plot of DECs for POAF compared with SR in GSE97455. (A) The heatmap displays top 20 up- and down-regulated DECs, Red and blue represent up- and down- expression. (B) The volcano plot showed all DECs for POAF. Red and green triangles refer to the significant DECs based on the selection criteria (Fold change > 2 and P value < 0.05). DELs, differentially expressed circRNAs; POAF, post-operative atrial fibrillation; SR, sinus rhythm.

Figure 8

CeRNA network construction. (A-D) The Venn diagrams show the intersection of predicted miRNAs targeting mRNAs (C1QA, C1R, MET, SDC4) using four databases (miRDB, miRWalk, RNA22, RNAInter). (E) The constructed ceRNA network includes three mRNAs, eight predicted miRNAs, and one circRNAs. ceRNA, competing endogenous RNA; C1QA, Complement C1q A Chain; C1R, Complement C1r; MET, MET Proto-Oncogene, Receptor Tyrosine Kinase; SDC4, Syndecan 4.

Figure 9

Immune cell infiltration between POAF and SR. (A) The proportion of 21 subtypes of immune cells in different samples regarding POAF and SR groups. (B) The barplot shows the comparison of 21 immune cell subtypes proportion between POAF and SR groups. Red and blue column represent SR and POAF group, respectively. *P < 0.05. (C) Correlation matrix of all 21 immune cell subtype compositions. The correlation coefficients are shown in the corresponding grids.