Investigation of the use of urinary proteomics in the prediction of childhood asthma.

PURPOSE: Urinary proteomics may help us improve our understanding and classification of asthma. We sought to identify clusters of children based on urinary protein profiles and explore associations between these clusters, specific proteins, and asthma. METHODS: We analyzed urine samples from 146 children (Kindergarten-Grade 8) using tandem mass-spectrometry. The presence of 4,080 urinary proteins was assessed. Unsupervised k-means clustering (KMN) identified clusters. Ten notable proteins per cluster were identified. We then tested three asthma prediction approaches: (1) all proteins only, (2) all proteins and non-protein features (e.g., socio-demographics, clinical), and (3) clusters and non-protein features. Each approach was run using four supervised machine learning (ML) algorithms. Traditional binary logistic regression (traditional analysis) was also conducted. RESULTS: We identified two clusters from the KMN. Among the notable proteins, only collagen alpha-1 chain was statistically different between those with and without asthma (p = 0.02). Approach 3 (cluster and non-protein features) using random forest methods had the best predictive performance. While cluster was listed as an important variable within Approach 3, it was not associated with asthma based on traditional analysis. Wheeze and mother's history of asthma were consistently associated with asthma. Unique predictors for the ML included %predicted FEF25-75%, %predicted FEV1/FVC, child's age, and mother's history of allergy; whereas, urban residence was unique to traditional analysis. CONCLUSION: We identified collagen alpha-1 chain as a potential biomarker for asthma. External validation in independent cohorts and stratification by clinical phenotypes are needed to confirm its diagnostic utility and clarify its role in asthma pathophysiology.