Integration of transcriptomics and machine learning to explore inflammatory characteristics and key oxidative
Yuelong Gu1,2,3, Ru Tang1,2,3, Zhihan Liu1,2,3
1Department of Otolaryngology-Head and Neck Surgery & Allergy Center, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Background:
Chronic rhinosinusitis with nasal polyps (CRSwNP) is a chronic inflammatory disease of the upper airway driven by diverse inflammatory cells and mediators. This study aimed to characterize the inflammatory landscape of CRSwNP and to identify and validate oxidative stress-related key genes in CRSwNP.
Methods:
RNA sequencing of nasal mucosal tissues was performed in 22 subjects, including 8 controls, 5 patients with non-eosinophilic CRSwNP (neCRSwNP), and 9 patients with eCRSwNP. Differential expression analysis was conducted using DESeq2 R package. Enrichment analysis, Ingenuity Pathway Analysis (IPA),weighted gene co-expression network analysis (WGCNA) and CIBERSORT was performed. Differentially expressed genes (DEGs), eosinophilic inflammation-associated module genes, and oxidative stress-related genes were intersected to generate candidate genes, followed by protein-protein interaction (PPI) network analysis. Six machine learning models were constructed using the candidate genes. SHapley Additive exPlanations (SHAP) were used to interpret feature contributions, and receiver operating characteristic (ROC) curves were used to assess discriminatory performance. External single-cell transcriptomic datasets and nasal tissue samples were used for validation.
Results:
Enrichment analyses showed that eCRSwNP was prominently associated with immune-inflammatory responses and oxidative stress-related pathways. WGCNA identified four modules closely related to eosinophilic inflammation, and five oxidative stress-related core signature genes were ultimately selected: HIF1A, RAC2, SELP, NCF2, and NCF4. ROC analysis demonstrated good discriminatory ability for all five genes. SHAP analyses indicated consistent directions of effects and feature-prioritization patterns across algorithms. External validation and nasal tissue sample experiments confirmed the overall upregulation of these genes in eCRSwNP. Cell-type localization analyses indicated that NCF2, NCF4, and RAC2 were mainly derived from myeloid cells, and SELP was predominantly localized to vascular endothelial cells, and HIF1A was broadly expressed within inflammatory cell infiltration area.
Conclusion:
By integrating transcriptomic analysis with multi-algorithm machine learning, we identified and validated HIF1A, RAC2, SELP, NCF2, and NCF4 as oxidative stress-related core signature genes in eCRSwNP and confirmed their elevated expression and cell-type-specific localization in eCRSwNP tissues. These genes may contribute to reactive oxygen species-associated inflammatory processes and represent candidate molecular signatures for eCRSwNP endotyping and further mechanistic investigation.
