Integrative machine learning identifies robust inflammation-related diagnostic biomarkers and stratifies immune-heterogeneous subtypes in Kawasaki disease.

BACKGROUND

Kawasaki disease (KD), a pediatric systemic vasculitis, lacks reliable diagnostic biomarkers and exhibits immune heterogeneity, complicating clinical management. Current therapies face challenges in targeting specific immune pathways and predicting treatment responses.

METHODS

Multi-cohort transcriptomic data were integrated to identify inflammation-related genes (IRGs). Differential analysis, weighted gene co-expression network analysis (WGCNA), and machine learning algorithms (LASSO, Boruta, SVM-RFE, Random Forest) were applied to screen diagnostic biomarkers. Immune infiltration and molecular subtyping based on diagnostic biomarkers were analyzed, complemented by regulatory network analysis to explore transcriptional, pharmacological, and miRNA interactions.

RESULTS

Six robust diagnostic biomarkers (ADM, ALPL, FCGR1A, HP, S100A12, SLC22A4) were identified, achieving AUC > 0.9 in cohorts. KD exhibited elevated neutrophils, monocytes, and Tregs but reduced CD8 + T cells and cytolytic activity. Consensus clustering stratified KD into two immune-heterogeneous subtypes: Cluster1 (neutrophil/Treg-dominant, enriched in TLR signaling) and Cluster2 (B cell/CD8 + T cell-dominant, linked to cytolytic activity). Regulatory networks revealed subtype-specific transcriptional regulators and therapeutic agents.

CONCLUSION

This study establishes inflammation-related diagnostic biomarkers and immune-stratified subtypes for KD, offering a framework for precision immunomodulatory therapies.