BACKGROUND: Pediatric sepsis is a life-threatening condition characterized by a dysregulated immune response to infection, often involving heightened oxidative stress. Understanding the molecular heterogeneity of sepsis can provide insights into potential therapeutic targets and diagnostic biomarkers. METHODS: Machine learning approaches were employed to identify diagnostic biomarkers. Unsupervised clustering was performed to identify distinct sepsis subtypes. We conducted an integrative analysis combining Gene Set Variation Analysis (GSVA), Gene Set Enrichment Analysis (GSEA), differential gene expression, and functional enrichment to study oxidative stress-related subgroups in sepsis patients. Immune cell infiltration and immune-related pathway activities were analyzed using the ssGSEA algorithm. GSVA and GSEA indicated significant enrichment of oxidative stress-related pathways in sepsis patients compared to controls. RESULTS: Differential expression analysis identified 371 upregulated and 304 downregulated genes in sepsis, with 34 genes linked to oxidative stress. LASSO and Random Forest analyses highlighted key diagnostic genes (GBA and MGST1), validated in independent datasets (GSE13904) with high diagnostic accuracy (AUC: GBA = 0.924, MGST1 = 0.857). Unsupervised clustering revealed two distinct sepsis subtypes with differential immune cell infiltration and pathway activities: Subtype 1 showed higher T cell and TFH infiltration, while Subtype 2 exhibited increased macrophage infiltration. Functional enrichment and GSEA identified key metabolic, oxidative stress, and immune pathways that were enriched in Subtype 2. CONCLUSION: Our comprehensive bioinformatics analysis unveils significant oxidative stress-related molecular heterogeneity in sepsis, identifying potential diagnostic biomarkers and therapeutic targets. Personalized medicine approaches targeting specific oxidative stress pathways and immune responses could enhance sepsis management and patient outcomes.