A Machine Learning and Single-Cell Derived Ferroptosis Signature for Diagnosis and Infiltration in Rheumatoid Arthritis

RA transcriptomic data and ferroptosis-related genes (FRGs) from FerrDb were integrated. WGCNA, consensus clustering, and machine learning algorithms were used to screen hub ferroptosis-related differentially expressed genes (hub RA-FRDEGs). An independent cohort and qPCR were employed to validate diagnostic value.

Upstream miRNAs and transcription factors were predicted. The immune microenvironment was assessed by ssGSEA, and biological functions were elucidated by single-gene GSEA. Drug prediction, molecular docking, and single-cell RNA-seq were performed to verify cellular localization.Results A total of 24 ferroptosis-related differentially expressed genes (RA-FRDEGs) were identified.

Consensus clustering stratified patients into three molecular subtypes (C1-C3), showing significant heterogeneity. Five machine learning algorithms prioritized four hub genes (GDF15, SLAMF8, AIM2, NTRK2), with AUC >0.8 in an independent validation cohort and qRT-PCR. ssGSEA revealed significant elevation of 11 immune cell types in RA samples, correlating with hub gene expression. Four potential therapeutic agents were identified through drug prediction and molecular docking.

Single-cell RNA-seq revealed intercellular communication features.Conclusion This study identifies four ferroptosis-related key genes (GDF15, SLAMF8, AIM2, NTRK2) in RA, which demonstrate significant diagnostic potential. These findings provide promising biomarkers for early diagnosis and new directions for targeted therapy in RA.