Screening of oxidative stress-related signature genes and construction of an interpretable discriminant model for preeclampsia

Background Preeclampsia (PE) is a serious pregnancy complication threatening maternal and fetal health, and its onset and progression are closely associated with placental oxidative stress. However, a systematic understanding of oxidative stress-related genes in PE and their roles in immune regulation remains lacking. Objective To identify oxidative stress-related signature genes in preeclampsia, validate their expression and tissue localization in placental tissues, construct an interpretable machine learning discriminant model, and explore their relationship with placental immune infiltration. Methods Nine GEO datasets were classified into three groups: a feature gene screening group (GSE10588, GSE75010), an artificial neural network training group (GSE30186, GSE44711, GSE54618, GSE24129, GSE43942, GSE147776), and an independent validation group (GSE25906). Differentially expressed genes (DEGs) were identified by limma, and their intersection with oxidative stress-related genes yielded candidate DEOSRGs. Multi-algorithm recursive feature elimination (SVM-RFE, RF-RFE, XGB-RFE) with five-fold cross-validation selected the core signature genes. Placental protein expression and trophoblast localization were assessed in clinical specimens (50 PE vs. 50 controls). An artificial neural network (ANN) discriminant model was constructed on the training group, with GSE25906 used for independent validation and SHAP-based interpretability analysis. Immune infiltration was analyzed by CIBERSORT on GSE75010. Results We identified 87 DEGs (65 upregulated, 22 downregulated) and 17 candidate DEOSRGs. Consensus recursive feature elimination screening yielded five core signature genes: LEP, HK2, NDRG1, membrane metallo-endopeptidase, and SERPINA3. Western blotting confirmed significant upregulation of all five proteins in PE placental tissues (P < 0.05), and immunohistochemistry localized positive signals predominantly to placental villous syncytiotrophoblasts. The ANN model achieved a 5-fold cross-validation AUC of 0.888 ± 0.010 and maintained 0.819 ± 0.010 in GSE25906, with LEP contributing most to classification. Immune infiltration profiling showed these genes were closely correlated with alterations in monocytes, M2 macrophages, Tregs, and neutrophils. Conclusion Five oxidative stress-related signature genes (LEP, HK2, NDRG1, MME, and SERPINA3) were identified and found significantly upregulated in PE placental tissues. An interpretable neural network discriminant model was constructed and independently validated. These biomarkers reflect placental pathophysiological alterations and may contribute to PE progression through coordinated modulation of oxidative stress and the maternal–fetal immune microenvironment.