Identification of ferroptosis biomarkers and immune infiltration landscapes in atrial fibrillation: A bioinformatics analysis

Abstract

Ferroptosis has been recognized as a critical factor in the development of atrial fibrillation (AF), but its precise mechanisms remain unclear. We downloaded the GSE115574 dataset from the gene expression omnibus database to analyze the expression levels of ferroptosis-related genes (FRGs) and identify differentially expressed genes (DEGs). Least absolute shrinkage and selection operator (LASSO) and support vector machine-recursive feature elimination (SVM-RFE) machine learning techniques were employed to identify key genes associated with AF. The diagnostic performance of these genes was evaluated using Receiver operating characteristic curves (ROC) and validated in an independent AF dataset. miRNA and lncRNA predictions for potential binding to these key genes were conducted using miRBase, miRDB, and TargetScan. Furthermore, gene set enrichment analysis (GSEA) enrichment analysis, immune cell infiltration analysis, and targeted drug prediction were performed. The intersection of LASSO regression and SVM-RFE analyses identified 7 DEGs significantly associated with AF. Validation through ROC and an additional dataset confirmed the importance of MAPK14, CAV1, and ADAM23. Significant infiltration of memory B cells, regulatory T cells, and monocytes was observed in atrial tissues. Seventy-two miRNAs were predicted to potentially target MAPK14, and 2 drugs were identified as targeting CAV1. This study underscores the involvement of FRGs in AF through machine learning and validation approaches. The observed immune cell infiltration suggests a potential link between immune response and AF. The predicted ceRNA network offers new insights into gene regulation, presenting potential biomarkers and therapeutic targets for AF.

Figure 1.

The overall protocol of this study.

Figure 2.

The gene differential expression in atrial fibrillation. (A) The heatmap of 73 FRGs in GSE115574. (B) Correlation between the expression of 73 FRGs. (C–E) The main GO terms and KEGG pathways enriched by differentially expressed mRNAs.

Figure 3.

Getting key FRGs through machine learning. (A and B) The performance in of ten-time cross-verification for tuning parameter in selection least absolute shrinkage and selection operator (LASSO). Each coefficient curve in the upper picture represents a single gene. The solid vertical lines in another picture represent the partial likelihood deviance SE, and the number of genes (n = 10) corresponding to the lowest point of the curve is the most suitable for LASSO. (C and D) Accuracy graph and cross-validation error graph of FRGs screened with the SVM-RFE algorithm. (E) Venn diagram of the intersection of FRGs screened with the 2 machine learning methods.

Figure 4.

ROC curves and GSEA results for groups. (A) ROC curves of the 7 FRGs. (B) Logistic regression model (AUC value was 0.957). (C) Plots of GSEA results for groups with high expression of FRGs.

Figure 5.

Differential immune cell infiltration in AF. (A) Differences in the infiltration of 22 types of immune cells in AF. Treat is AF. (B) Heat map of the correlation between the expression levels of the 7 FRGs and the infiltrating numbers of 22 types of immune cells in each group. Red squares: positive correlation; purple squares: negative correlation. Darker color indicates stronger correlation. Not significant (ns), *P < .05, **P < .001, ***P < .001.

Figure 6.

Construction of drug regulatory networks and ceRNA networks. (A) The GSVA results of the metabolic pathways related to the 7 FRGs. (B) Targeted drug prediction. (C) Construct ceRNAs for ADAM23, CAV1, and MAPK14.