Precision Breathing: Asthma Phenotyping via Machine Learning

Asthma is a complex condition characterized by chronic airway inflammation, with varying severity, symptoms, triggers, and treatment responses. Traditional classification relies on clinical attributes, but the growing understanding of asthmas heterogeneity highlights the need for phenotyping. Effective management requires regular monitoring, medication, and prevention of exacerbations, but current diagnostic methods face challenges such as the lack of definitive tests and reliance on subjective measures. Implementing precision medicine, especially for severe cases, necessitates identifying measurable markers in biofluids. This study explores machine learning methods to identify biomarkers differentiating various asthma phenotypic states. We measured inflammatory markers in both plasma and saliva samples and used machine learning algorithms to determine their efficacy in reflecting airway inflammation. Our findings indicate that saliva markers provide a more accurate representation of localized inflammation compared to plasma markers, which reflect a systemic response. Using MRMR (Minimum Redundancy Maximum Relevance) ranking, we enhanced model efficacy. The K-Nearest Neighbor (KNN) classifier achieved 75% accuracy with the first 12 saliva markers, while the Random Forest (RF) classifier performed best for plasma models, though with lower accuracy. Our results suggest machine learning can effectively identify key markers for asthma phenotyping, aiding personalized treatment strategies. Customizable point-of-care devices could validate these models and improve their accuracy, advancing asthma treatment and management.