Integrating disulfidptosis patterns and machine learning identifies a novel diagnostic signature for endometriosis

Endometriosis (EM) is a heterogeneous gynecological disorder with incompletely understood molecular mechanisms. Disulfidptosis, a recently identified form of programmed cell death, has not been systematically explored in EM. Gene expression profiles from the GSE51981 dataset were analyzed to investigate disulfidptosis-related genes (DRGs) and immune characteristics in EM. Consensus clustering identified molecular subtypes, and weighted gene co-expression network analysis (WGCNA) was performed to screen subtype-associated genes. Four machine-learning algorithms were subsequently compared to establish an optimal diagnostic model, which was further evaluated in two independent external datasets and validated at the expression level using RT-qPCR experiments. EM tissues showed abnormal DRG expression patterns and immune infiltration. Two disulfidptosis-associated molecular subtypes with distinct immune features and biological pathways were identified. The Extreme Gradient Boosting (XGB) algorithm showed the best performance and generated a five-gene signature comprising BTBD2 , SLC4A2 , SSC5D , LPPR2 , and SRGAP2C . External validation, calibration, decision curve analysis, and nomogram evaluation provided complementary evidence supporting the potential diagnostic utility of the signature. RT-qPCR provided partial expression-level validation of selected genes within the five-gene signature in EM patients. Disulfidptosis-related molecular heterogeneity is associated with immune remodeling in EM. The five-gene signature provides a potential diagnostic tool and new insights into EM mechanisms.