Identification and validation of a robust inflammation-related three-gene signature for the diagnosis of neonatal necrotizing enterocolitis via machine learning integration.

PURPOSE: Neonatal necrotizing enterocolitis (NEC) is a severe gastroesophageal emergency in neonates, especially premature infants, with acute inflammation and necrosis. Due to the nonspecific nature of early clinical manifestations, delayed diagnosis remains an issue often associated with high mortality and morbidity, hence finding molecular biomarkers with high diagnostic specificity is critical for better clinical outcomes. This work combines transcriptomic data with machine learning algorithms to identify a diagnostic signature that strongly correlates to NEC. METHODS: The GSE64801 transcriptome dataset from GEO was trained and the independent GSE46619 dataset was validated externally. A bioinformatics pipeline was developed to detect differentially expressed genes (DEG) between NEC and control tissues intersected with 935 known inflammation-related genes (IRGs) to identify inflammation-specific candidates. For feature robustness, the combinatorial screening strategy involving least absolute shrinkage and selection operator regression and Boruta random forest algorithm was used. Diagnostic performance of identified core biomarkers was validated across cohorts. A diagnostic nomogram was constructed and performance was evaluated using Receiver Operating Characteristics (ROC) curves, calibration plots, Decision Curve Analysis (DCA). Various downstream analyses including GSEA, regulatory network construction, drug target prediction were performed. RESULTS: In the training cohort 452 DEGs were identified, down to 20 candidates after intersecting with IRGs. The two machine learning strategies identified three key biomarkers: REG3A, CXCL9 and Kininogen 1 (KNG1). Validation assays showed that these genes are upregulated in NEC tissues in both cohorts. The nomogram constructed from this three genes was more diagnostically effective and achieved Area Under the Curve (AUC) of 0.956 in the training cohort. Mechanistically GSEA plays a significant role in key immune-inflammatory processes such as chemokine signaling, intestinal immune networks and graft-versus-host responses. Further upstream transcriptional regulation network was constructed and potential therapeutic agents such as avoralstat were predicted. CONCLUSION: We identified and validated a novel three gene inflammatory signature (REG3A, CXCL9, KNG1) high diagnostic accuracy for NEC. These biomarkers are involved in the basic pathophysiology of the disease, providing a scientific basis for early diagnostic tools and precise treatments.