Identification of Immune-Related Hub Genes as Candidate Biomarkers for Preeclampsia: Integrative Transcriptomic and Machine Learning Approaches.

Abstract

Preeclampsia (PE) is a major cause of maternal and perinatal morbidity and is increasingly recognized as a disorder involving immune dysregulation. However, reliable candidate biomarkers and key molecular alterations associated with PE remain unclear. We integrated multiple public placental transcriptomic datasets from patients with PE and normotensive controls. Weighted gene co-expression network analysis and eight machine-learning algorithms were used to identify stable immune-related candidate genes, followed by immune infiltration analysis, single-cell analysis, SHAP interpretation, and in silico gene knockout analysis. A six-gene logistic regression model was evaluated in the same merged placental dataset used for feature selection and explored in a small independent peripheral-blood cohort. Frequency-based integration across the eight algorithms identified a six-gene signature comprising ACSS3, F13A1, MCTP1, PTGFR, SLC16A10, and TNFRSF11A. The model showed apparent discrimination in the merged placental dataset (AUC = 0.869, 95% CI: 0.820-0.919) and exploratory cross-tissue discrimination in the peripheral-blood cohort (AUC = 0.860, 95% CI: 0.691-1.000). Immunohistochemical analysis of placental tissues from five women with PE and five normotensive controls showed lower protein abundance of all six candidates in PE, consistent with the bulk transcriptomic findings. Single-cell analysis indicated heterogeneous cell-type-dependent expression patterns, while in silico knockout analysis suggested potential macrophage-associated regulatory effects. Collectively, these exploratory findings identify a six-gene immune-related candidate signature associated with PE that requires validation in larger, prospectively collected cohorts before clinical application.