Identification of Glycolysis-Related Signature and Molecular Subtypes in Child Sepsis Through Machine Learning and Consensus Clustering: Implications for Diagnosis and Therapeutics.

Abstract

Pediatric sepsis remains one of the leading causes of mortality in children worldwide. Despite advances in medical care, the prognosis of pediatric sepsis is still poor, necessitating the need for more precise diagnostic and therapeutic strategies. Recently, metabolic reprogramming, particularly glycolysis, has been implicated in the pathogenesis of sepsis, offering new avenues for biomarker discovery and targeted therapy. We applied the GSVA algorithm to the GSE26440 dataset to score glycolysis pathways and identified key glycolysis-related genes (GRGs) using LASSO and logistic regression. We then constructed a predictive nomogram with these GRGs and used consensus clustering to define new molecular subgroups, followed by analyzing their metabolic and immune characteristics. The signature genes were validated by animal experiments. We found increased glycolysis pathway activity in sepsis patients. Through the application of LASSO and logistic regression, GNPDA2, PRKACB, and TGFBI emerged as potential glycolysis-based diagnostic markers. The nomogram showed significant diagnostic accuracy in both the original (GSE26440) and the separate validation datasets (GSE13904 and GSE26378). We distinguished two sepsis subtypes, with the C2 subtype exhibiting higher GRGs, glucose metabolism, and inflammation. Immune infiltration and checkpoint gene expression also varied between the subtypes. Our research identifies glycolysis-based diagnostic markers and molecular subtypes in sepsis, enhancing our understanding and potentially leading to better diagnosis and treatment strategies, including immunotherapy.