Machine learning reveals immune heterogeneity in recovery-phase samples of pediatric wheezing.

Background: Pediatric wheezing diseases are linked to the risk of progression to persistent wheezing or asthma, particularly in cases with chronic airway inflammation and genetic predisposition. Early identification of high-risk children is crucial for timely intervention, yet the recovery phase of wheezing episodes, a transitional period, is often overlooked. Methods: We integrated RNA-seq data from GEO datasets (GSE113209 and GSE103166), focusing on 39 recovery-phase nasal mucosal samples from two independent cohorts. After log-transformation and batch-effect correction using ComBat, differential expression analysis was performed using the limma-trend method. Hierarchical clustering identified two distinct clusters, and differential gene expression (DEGs) was analyzed using GO, KEGG, GSEA, and DO methods. Immune infiltration patterns were assessed with CIBERSORT. Machine learning approaches (lasso regression, SVM-RFE) identified key biomarkers for potential subgroups in recovery samples. Results: Clustering revealed two clusters with differential immune infiltration patterns. Cluster 2 exhibited increased eosinophil, mast cell, and memory B cell infiltration. DEGs included 791 upregulated and 517 downregulated genes (|log2FC| >0.25, Conclusions: Recovery-phase samples of wheezing illness exhibit heterogeneity. Cluster analysis revealed a subset demonstrating propensity for pro-inflammation, while biomarkers identified through machine learning correlated with immune infiltration. Furthermore, using external cohort, we confirmed the association between IL18R1 expression in recovery-phase samples and recurrent wheezing episodes. The pro-inflammatory subpopulation identified in recovery-phase samples warrants further investigation.