# Identification and analysis of exosome-associated signatures in pediatric sepsis by integrated bioinformatics analysis and machine learning

## TL;DR

This study identifies exosome-related genes linked to pediatric sepsis and develops accurate machine learning models for early diagnosis.

## Contribution

The novel contribution is the integration of bioinformatics and machine learning to identify exosome-associated signatures and robust biomarkers for pediatric sepsis.

## Key findings

- 21 exosome-related genes showed significant differential expression in pediatric sepsis.
- Machine learning models achieved high diagnostic accuracy (AUC > 0.995) using key biomarkers like CD177, GYG1, IRAK3, MCEMP1, and TLR5.
- Immune-related pathways such as phagocytosis and NF-κB signaling were enriched in exosome-related gene networks.

## Abstract

Pediatric sepsis (PS) is a critical condition characterized by life-threatening organ dysfunction and immune dysregulation, including exosome-mediated immune modulation, often linked to infections. Investigating the role of exosome-related genes (ERGs) in the pathogenesis of PS is essential for identifying significant diagnostic and therapeutic targets.

Four datasets, namely GSE66099 (training set) and GSE13904, GSE26378, and GSE26440 (validation sets), were retrieved from the Gene Expression Omnibus (GEO). The differential expression of 56 ERGs was analyzed, followed by consensus clustering to identify distinct exosome-related patterns in PS. Weighted gene co-expression network analysis (WGCNA) was utilized to identify PS-related genes (SRGs). Additionally, the immune microenvironment was assessed, and diagnostic models were developed employing specific machine learning algorithms.

The differential expression analysis identified 21 ERGs that exhibited significant alterations in PS. Consensus clustering revealed two distinct subtypes of PS based on the expression pattern of ERGs. WGCNA identified several hub genes involved in exosome function and PS, with immune-related pathways, including phagocytosis and NF-κB signaling, showing significant enrichment. These genes were leveraged to construct machine learning models, which demonstrated a high diagnostic accuracy, with an area under the curve (AUC) > 0.995. The analysis identified CD177, GYG1, IRAK3, MCEMP1, and TLR5 as key biomarkers. Furthermore, external validation confirmed the superior performance of the constructed model.

This study elucidated the role of ERGs in PS, and highlights the significance of immune dysregulation in the pathogenesis of the disease. The developed diagnostic models represent promising tools for the early detection and prognosis prognostic of PS.