Machine Learning and Functional Analysis of Placental DNA Methylation Signatures for Early-Onset Preeclampsia

Early-onset preeclampsia (EOPE) is a pregnancy disorder characterized by abnormal placental development and substantial maternal and fetal risk. DNA methylation at CpG sites has emerged as a key epigenetic mechanism in EOPE. In this study, we developed machine learning models for EOPE classification using β-values from CpG sites in placental tissue. Starting from 599 EOPE-associated CpG sites reported earlier, we used attribute selection in WEKA to identify 30 statistically significant descriptors, which mapped to 19 unique genes. Multiple classifiers were trained and evaluated, and the SPegasos classifier demonstrated the best performance, achieving 95.45% accuracy and an area under the receiver operating characteristic curve of 0.9545 in tenfold cross‑validation, and 95% accuracy in an independent validation dataset. These findings suggest a potentially high correlation between the CpG methylation signatures and EOPE. By creating a machine learning model, we can potentially classify a sample as having EOPE or a normal pregnancy using DNA methylation data at specific CpG sites in placental tissue. Furthermore, we identified the adipogenesis pathway to be central to the disease's pathophysiology. Gain ratio analysis highlighted a subset of CpG sites with particularly strong discriminative power between EOPE and control placentas. Functional annotation of the 19 CpG‑associated genes using protein and pathway enrichment analyses revealed enrichment in signal transduction, steroid and lipid metabolism, and cell developmental processes. Pathway analysis identified adipogenesis as a central pathway, consistent with dysregulated adipokine signaling and systemic inflammation in EOPE. These findings demonstrate that placental CpG methylation signatures can accurately distinguish EOPE from normal pregnancies and provide mechanistic insight into disrupted adipogenesis‑related pathways. The identified CpG methylation signature may serve as a basis for developing early molecular biomarkers and targeted strategies to improve EOPE diagnosis and management.