Computational discovery of PGD, MAPK14, and KRAS as diagnostic biomarkers for neonatal sepsis through integrated machine learning, immune infiltration analysis, and molecular docking

BackgroundNeonatal sepsis is a life-threatening condition with high mortality. Ferroptosis and cuproptosis, oxidative stress-related cell death pathways, are implicated in sepsis pathogenesis, but their role in neonatal sepsis remains unclear.

This study aimed to identify and validate diagnostic biomarkers for neonatal sepsis associated with ferroptosis and cuproptosis pathways using integrated bioinformatics and machine learning approaches, and to explore potential therapeutic targets.MethodsTranscriptomic data from neonatal sepsis patients (GSE69686, GSE25504) were analyzed. Differential expression analysis, weighted gene co−expression network analysis (WGCNA), and protein−protein interaction (PPI) networks were performed to identify cuproptosis− and ferroptosis−related genes (CFRGs).

Three machine learning algorithms—LASSO, SVM−RFE, and XGBoost—were applied for feature selection. Immune infiltration was assessed via CIBERSORT.

Molecular docking was used to screen FDA−approved drugs against candidate targets. In vitro validation was conducted using LPS−stimulated THP−1−derived macrophages, with gene expression measured by RT−qPCR and drug effects assessed by CCK−8 assay.ResultsThree biomarkers—PGD, MAPK14, and KRAS—were consistently identified by all machine learning models and showed strong diagnostic performance (AUC > 0.79 in external validation).

Immune infiltration analysis revealed increased neutrophils and Tregs, and decreased CD8+ T cells in sepsis.