The Landscape of Disulfidptosis in Preeclampsia Reveals a Novel 5-Gene Diagnostic Signature via Machine Learning

Preeclampsia (PE), a pregnancy-specific pathological condition, has shown a growing incidence over recent decades. Disulfidptosis is a newly discovered mode of programmed cell death that differs from traditional cell death pathways in its molecular mechanisms. Numerous studies have reported the association between disulfidptosis and various diseases; however, the role of disulfidptosis in the pathogenesis of PE remains unknown.

This study first analyzed the expression patterns of disulfidptosis-related genes (DRGs) using the GSE75010 dataset. Based on these results, unsupervised consensus clustering was conducted specifically on the PE samples included in this dataset. Weighted gene co-expression network analysis and machine learning algorithms were utilized to identify hub genes related to PE and disulfidptosis clusters. Ultimately, the expression profiles of these hub genes were validated using the independent datasets GSE4707, GSE30186, and GSE54618, as well as quantitative PCR (qPCR).

9 DRGs showed abnormal expressions in the PE samples (P < 0.05). Subsequently, two disulfidptosis clusters were identified, each with its own unique functional pathway. Among the four algorithms examined, the support vector machine delivered the most dependable predictions (sensitivity = 0.889). In addition, the genes SASH1, CST6, CCBL1, FSTL3, and SPAG4 were determined as the central genes. A diagnostic model was established using these five genes, The high-fitting of the calibration curve and the positive net benefit of the broad threshold of the DCA decision curve demonstrate that this model has potential good clinical decision-making value. Meanwhile, the receiver operating characteristic (ROC) analysis of the external validation dataset indicates that the predictive performance of this model is excellent (AUC =1), the qPCR validation demonstrated that the differences in expression of the five genes in the PE and non-PE controls (NC) samples were statistically significant (P < 0.05). However, the area under the curve (AUC) values of the external validation all being equal to 1 may indicate a problem of overfitting. The sample size verified by qPCR is relatively small, which has statistical limitations.

This study proposes a new diagnostic model for PE, which can serve as a framework for studying disease heterogeneity and provides a basis for understanding the role of disulfidptosis in the occurrence of PE.