Machine Learning and Bioinformatics Approaches to Identify the Candidate Biomarkers in Severe Asthma

Severe asthma is characterized by a poor level of control that severely affects the patient's life and prognosis. However, the underlying pathogenic mechanisms remain unknown. Here, we identified differentially expressed genes from the microarray datasets(GSE130499 and GSE63142) of severe asthma, and then constructed models to screen the most relevant biomarkers to severe asthma by machine learning algorithms(LASSO and SVM-RFE), with further validation of the results by GSE43696. Three genes (BCL3, DDIT4 and S100A14) are considered as biomarkers of severe asthma and had good diagnostic effect. Among them, BCL3 transcript level was down-regulated in severe asthma, while S100A14 and DDIT4 transcript levels were up-regulated. Next, the features of the immune microenvironment in severe asthma were analyzed and single-cell datasets(GSE193816 and GSE227744) were identified for potential biomarker-specific expression and intercellular communication. Infiltration of neutrophils and mast cells were found to be increased in severe asthma and may be associated with bronchial epithelial cells through BMP and NRG signaling. Finally, The expression levels of potential biomarkers were verified with a mouse model of asthma.