Abstract
BACKGROUND: Rheumatoid arthritis (RA) has complex pathological mechanisms, and mitochondria are critical to its occurrence and development while the specific mechanisms remain unclear. This study aimed to identify key mitochondria-related genes in RA via machine learning, validate them by two-sample Mendelian randomization (MR), and provide novel therapeutic target clues. METHODS: RA-associated gene microarray datasets from GEO were curated by R software. Differential expression profiling, LASSO regression, SVM classification and RF modeling were used to screen hub genes and build diagnostic models. GO, KEGG and Metascape analyses clarified their biological roles, and immune infiltration profiling and consensus clustering were performed for immune heterogeneity analysis. MR analysis identified core genes linked to RA susceptibility, followed by drug enrichment and molecular docking. A preliminary RT-qPCR validation was conducted on synovial tissues from a small cohort of 3 RA patients and 3 controls. RESULTS: A total of 1,432 differentially expressed genes (DEGs) were identified, and six mitochondria-related hub genes (UCP2, BCL2A1, FASN, AKR1B10, IFI27, PDK1) were screened by machine learning. The diagnostic model based on these genes had good discriminatory ability. MR analysis confirmed a causal relationship between BCL2A1 and RA, suggesting it as a potential risk factor. Drug enrichment and molecular docking showed these hub genes could bind to candidate drugs, and the exploratory RT-qPCR assay initially validated the bioinformatics findings. Leave-one-out sensitivity analysis confirmed that the RT-qPCR results exhibited good stability. CONCLUSION: Bioinformatics analysis identified UCP2, BCL2A1, FASN, AKR1B10, IFI27, and PDK1 as potential mitochondria-related diagnostic biomarkers for RA and established a corresponding diagnostic model. MR analysis showed a causal relationship between BCL2A1 and RA, indicating that BCL2A1 serves as a risk factor for the disease.