Abstract
BACKGROUND: Endometriosis is often diagnosed late and presents significant challenges in clinical treatment. A comprehensive investigation of the cellular classification and composition of endometriosis is essential for studying its diagnosis and treatment. METHODS: This study utilized the Gene Expression Omnibus (GEO) public database and referenced single-cell RNA sequencing (scRNA-seq) atlases. The CIBERSORTx algorithm was applied to perform deconvolution on the samples and estimate the proportions of endometrial cell subtypes. A random forest model was constructed to predict the diagnosis of endometriosis. Additionally, immunohistochemical validation was performed on the marker genes of MUC5B+ epithelial cells and dStromal late mesenchymal cells, which showed high diagnostic contribution. RESULTS: Endometriosis consists of 5 major cell types, further classified into 52 distinct cell subtypes. Compared to healthy controls, these subtypes exhibited varying degrees of alterations, with MUC5B+ epithelial cells, dStromal late mesenchymal cells, and M2 macrophages showing an increasing trend. Enriched signaling pathways were primarily associated with epithelial-mesenchymal transition (EMT), cell migration, and inflammatory responses. A random forest model, based on cell-type proportions, has been shown to achieve excellent diagnostic performance (AUC = 0.932), with MUC5B+ epithelial cells identified as the top predictive feature. Immunohistochemical validation confirmed high expression of the marker genes MUC5B and TFF3. CONCLUSION: By integrating single-cell and bulk transcriptomics, we identified MUC5B+ epithelial cells and dStromal-late mesenchymal cells as dual drivers of fibrosis and inflammation in endometriosis. Our findings revealed that MUC5B+ epithelial cells may serve as the top factor for the diagnosis of endometriosis.