Identification of key ferroptosis-related biomarkers in steroid-induced osteonecrosis of the femoral head based on machine learning

Abstract

BackgroundThis study was aimed to identify key ferroptosis-related biomarkers in steroid-induced osteonecrosis of the femoral head (SONFH) based on machine learning algorithm.MethodsThe SONFH dataset GSE123568 (including 30 SONFH patients and 10 controls) was used in this study. The differentially expressed genes (DEGs) were selected between SONFH and control groups, which were subjected to WGCNA. Ferroptosis-related genes were downloaded from FerrDb V2, which were then compared with DEGs and module genes. Two machine learning algorithms were utilized to identify key ferroptosis-related genes, and the underlying mechanisms were analyzed by GSEA. Correlation analysis between key ferroptosis-related genes and immune cells was analyzed by Spearman method. The drug–gene relationships were predicted in CTD.ResultsTotal 2030 DEGs were obtained. WGCNA identified two key modules and obtained 1561 module genes. Finally, 43 intersection genes were identified as disease-related ferroptosis-related genes. After LASSO regression and RFE-SVM algorithms, 4 intersection genes (AKT1S1, BACH1, MGST1 and SETD1B) were considered as key ferroptosis-related gene. The 4 genes were correlated with osteoclast differentiation pathway. Twenty immune cells with significant differences were obtained between the groups, and the 4 key ferroptosis-related genes were correlated with most immune cells. In CTD, 41 drug–gene relationship pairs were finally obtained.ConclusionsThe 4 key ferroptosis-related genes, AKT1S1, BACH1, MGST1 and SETD1B, were identified to play a critical role in SONFH progression through osteoclast differentiation and immunologic mechanisms. Additionally, all the 4 genes had good disease prediction effect and could act as biomarkers for the diagnosis and treatment of SONFH.