Journal Article

Multi-cohort transcriptomics integration for building and validating a diagnostic model of peripheral blood septic shock

Zhang, Yang, Li, Ling, Qian, Weiwei, Jiang, Hui, Zeng, Xia, Li, Kexun, Chen, Zhengjun, Peng, Hongqiong

2026

Overview

To integrate multi-cohort transcriptomic, single-cell, and experimental data to identify diagnostic signature genes for septic shock, establish a peripheral blood molecular diagnostic model, and elucidate the m6A regulatory mechanisms of key genes. Candidate genes were identified from five GEO peripheral blood cohorts through batch effect-corrected differential expression analysis and WGCNA, followed by parallel GO/DO enrichment analysis. Feature genes were selected using PPI networks combined with LASSO, SVM-RFE, and random forest algorithms. A 5-gene artificial neural network (ANN) diagnostic model was constructed and validated using ROC and logistic regression in GSE95233, GSE131761, and clinical cohorts. Immune cell composition and expression of characteristic genes in neutrophils were analyzed using CIBERSORT and GSE167363 single-cell data. The METTL14/YTHDF1-S100A12 m6A axis was elucidated via qRT-PCR, Western blot, MeRIP-qPCR, RIP-qPCR, and Actinomycin D experiments. In CLP mice, siMETTL14 was administered for intervention and assessment of lung injury. A total of 76 sepsis-shock-associated candidate genes were identified, enriched in the bacterial defense pathway. Five robust candidate genes (S100A12, MMP8, PGLYRP1, CEACAM8, MMP9) were selected by integrating PPI and three machine learning algorithms. The constructed ANN achieved high AUC across multiple cohorts, and all five genes showed significantly elevated mRNA and protein levels in peripheral blood from clinical sepsis patients. CIBERSORT and single-cell results indicated significant neutrophil expansion, with the five genes predominantly enriched in neutrophils and progressively elevated with worsening outcomes. m6A-related experiments demonstrated that METTL14 mediates m6A modification and stabilizes S100A12 mRNA through YTHDF1 recognition; knocking down either METTL14 or YTHDF1 accelerated its degradation. silencing of METTL14 in CLP mice reduced lung tissue injury and lipid peroxidation/ferroptosis levels. In this study, a diagnostic signature was established for septic shock comprising five neutrophil-associated genes and an ANN model, revealing the regulatory role of the METTL14/YTHDF1-mediated m6A-S100A12 axis in neutrophils. This suggests the METTL14/m6A pathway as a potential diagnostic and therapeutic target.

• Publisher: Frontiers Media SA,Frontiers Media S.A
• Subject: Algorithms / Animals / artificial neural network / Biomarkers / Cohort Studies / Datasets / diagnostic gene signature / Disease Models, Animal / Ferroptosis / Gelatinase B / Gene expression / Gene Expression Profiling / Humans / Leukocytes (neutrophilic) / Lipid peroxidation / Lungs / m6a modification / Machine learning / Mice / N6-methyladenosine / Neural networks / Neutrophil collagenase / Neutrophils / Neutrophils - immunology / Neutrophils - metabolism / Ontology / Original Research / Peripheral blood / peripheral blood transcriptome / Proteins / Regression analysis / RNA modification / Sepsis / Septic shock / Shock, Septic - blood / Shock, Septic - diagnosis / Shock, Septic - genetics / Therapeutic targets / Transcriptome / Transcriptomics