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

Abstract

AbstractBackground Severe asthma is defined as a persistent increase in airway inflammation despite the use of systemic glucocorticoids, targeted biologic therapies. Early prediction of severe asthma is challenging due to the lack of valuable biomarkers. The aim of this study was to identify crucial differentially expressed genes (DEGs) associated with severe asthma through approaches of bioinformatics analysis. Methods Three datasets GSE130499, GSE43142 and GSE43696 were derived from the GEO expression database. Two datasets (GSE130499 and GSE43142) were merged, and batch effects were removed by using the "SVA" package. Afterwards, the differentially expressed genes (DEGs) were analyzed with the "limma" package. Next, DEGs were functionally enriched and pathway analyzed in the online analysis website DAVID, then DEGs were analyzed again by two machine learning algorithms (LASSO and SVM-RFE) to obtain the candidate biomarkers, and the diagnostic validity of the biomarkers was assessed using subject operating characteristic (ROC) curves, and finally the results were further validated through the GSE43696. Results Total of 73 gene differential expression genes were identified in severe asthma and normal control. After screening with two machine learning algorithms, LASSO and SVM-RFE, three genes (BCL3,DDIT4andS100A14) were recognized as biomarkers of asthma and had good diagnostic effect. Among them,BCL3transcript level was down-regulated in severe asthma, whileS100A14andDDIT4transcript levels were up-regulated. Conclusions In this study, we identified three differentially expressed genes (BCL3,DDIT4andS100A14) of diagnostic significance that may be involved in the development of severe asthma and proposed new insights into the underlying mechanisms.