Deciphering the Immune Subtypes and Signature Genes: A Novel Approach Towards Diagnosing and Prognosticating Severe Asthma Through Interpretable Machine Learning

Abstract

Asthma, a pervasive pulmonary disorder, affects countless individuals globally. Characterized by chronic inflammation of the bronchial passages, its symptoms include cough, wheezing, dyspnea, and chest tightness. While many manage their symptoms through pharmaceutical interventions and self-care, a significant subset grapples with severe asthma, posing therapeutic challenges. This study delves into the intricate etiology of asthma, emphasizing the pivotal roles of immune cells such as T cells, eosinophils, and mast cells in its pathogenesis. The recent emergence of monoclonal antibodies, including mepolizumab, reslizumab, and benralizumab, offers therapeutic promise, yet their efficacy varies due to the heterogeneous nature of asthma. Recognizing the potential of personalized medicine, this research underscores the need for a comprehensive understanding of asthma’s immunological diversity. We employ single-sample gene set enrichment analysis (ssGSEA) and Least Absolute Shrinkage and Selection Operator (LASSO) algorithms to identify differentially expressed immune cells and utilize machine learning techniques, including Extreme Gradient Boosting (XGBoost) and random forest, to predict severe asthma outcomes and identify key genes associated with immune cells. Using a murine asthma model and an online database, we aim to elucidate distinct immune-centric asthma subtypes. This study seeks to provide novel insights into the diagnosis and classification of severe asthma through a transcriptomic lens.