The bronchial microbiome of patients with asthma differs from that of the healthy controls. Metagenomic analysis has been conducted using 16S recombinant DNA sequencing from extracellular vesicles (EVs) to identify microbiomes and provide culture-independent estimation of microbial diversity.1Lee J.H. Choi J.P. Yang J. et al.Metagenome analysis using serum extracellular vesicles identified distinct microbiota in asthmatics.Sci Rep. 2020; 10: 15125Crossref PubMed Scopus (5) Google Scholar Despite the construction of the metagenome of the asthmatic lung, limitations persist in sampling the bronchial airway. Exhaled breath condensate (EBC) contains aerosol and volatile compounds that can be analyzed to noninvasively understand the physiological and pathologic processes in the lung.2Hunt J. Exhaled breath condensate: an evolving tool for noninvasive evaluation of lung disease.J Allergy Clin Immunol. 2002; 110: 28-34Abstract Full Text Full Text PDF PubMed Scopus (270) Google Scholar This study performed metagenomic analysis using EBC-derived EVs to determine the characteristics of the microbiome in patients with asthma compared with those in healthy controls. Moreover, we suggested a diagnostic tool for asthma using artificial intelligence (AI) modeling based on the results of the microbiota composition in patients with asthma. The EBC was obtained from 58 healthy controls and 251 individuals with asthma at the Asan Medical Center between September 2014 and December 2019. Healthy controls were screened for respiratory diseases and recruited during regular wellness examinations. The eligibility criteria for patients with asthma were as follows: (1) patients older than 18 years; (2) symptoms such as dyspnea, wheezing, or cough for more than 3 months; and (3) airway hyperresponsiveness on a provocation test or airway reversibility after inhalation of a short-acting beta-agonist. Patients with severe lung damage, bronchiectasis, or a history of lung resection were excluded. Informed consent was obtained from all participants, and the study protocol was approved by the Institutional Review Board of Asan Medical Center (IRB No. 2006-0388). The EBC samples were collected using RTube (Respiratory Research, Austin, Texas). EVs were isolated from the EBC and analyzed. The extracted 16S recombinant DNA was subjected to next-generation sequencing as described previously.3Yang J. McDowell A. Kim E.K. et al.Consumption of a Leuconostoc holzapfelii-enriched synbiotic beverage alters the composition of the microbiota and microbial extracellular vesicles.Exp Mol Med. 2019; 51: 1-11Crossref Scopus (19) Google Scholar Taxonomic profiling was conducted for all samples at the genus level.4Yang J. Moon H.E. Park H.W. et al.Brain tumor diagnostic model and dietary effect based on extracellular vesicle microbiome data in serum.Exp Mol Med. 2020; 52: 1602-1613Crossref PubMed Scopus (3) Google Scholar Two algorithms of AI modeling including artificial neural network (ANN) and gradient boosting (GBM) were applied to selective EBC biomarkers.5Abadi M, Agarwal A, Barham P, et al. Tensorflow: large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467. Available at: https://arxiv.org/abs/1603.04467. Accessed March 16, 2016.Google Scholar,6Pedregosa F. Varoquaux G. Gramfort A. et al.Scikit-learn: machine learning in Python.J Mach Learn Res. 2011; 12: 2825-2830Google Scholar The data were split into testing and training sets 10 times before training to validate the performance of the models. Alpha diversity was ascertained using the phyloseq R package to calculate the observed species richness and Chao1 for the estimated species richness and the Shannon and Simpson indices for species evenness and richness with the rarefaction set at 1091. Beta diversity testing was performed using principal coordinate analysis for multidimensional scaling of the EBC samples. In addition, t tests were performed to evaluate the statistical difference between the control and asthma groups, in which P < .05 was considered statistically significant. The asthma group had significantly higher alpha diversity (P < .05) for all 4 measurements (Fig 1). The observed operational taxonomic units and Chao1 estimates revealed greater species richness in the asthma EBC samples. The Shannon and Simpson diversity indices both revealed higher alpha diversity measures in the individuals with asthma than those in the healthy controls. In the plot of principal coordinates analysis (based on Bray-Curtis distance of all samples of control and patients with asthma), 2 groups were clustered separately even though there were several patients with asthma in the same cluster of the control group. There was no difference in the airway microbiota related to airway inflammation according to the proportions of induced sputum eosinophils and neutrophils, severity and duration of asthma, obesity, smoking status, or use of inhaled corticosteroids. At the genus level, Sphingomonas, Akkermansia, Methylophaga, Acidocella, and Marinobacter were significantly more abundant in patients with asthma (P < .05). Based on the result of the microbiome-based differences between patients with asthma and controls, the diagnostic AI model for asthma using GBM and ANN revealed good performance with respective areas under the curve of 0.832 and 0.769. Firmicutes and Proteobacteria at the phylum level were common important features between the GBM and ANN asthma models. The change in microorganisms in the respiratory tract has been found to play an important role in mediating the pathogenesis of asthma.7Shukla S.D. Shastri M.D. Chong W.C. et al.Microbiome-focused asthma management strategies.Curr Opin Pharmacol. 2019; 46: 143-149Crossref PubMed Scopus (5) Google Scholar Our study identified bacterial compositional differences between patients with asthma and healthy controls using EBC-derived EVs, which is consistent with previous studies using induced sputum and bronchial washing.8Marri P.R. Stern D.A. Wright A.L. Billheimer D. Martinez F.D. Asthma-associated differences in microbial composition of induced sputum.J Allergy Clin Immunol. 2013; 131 (346-352.e1-3)Abstract Full Text Full Text PDF PubMed Scopus (243) Google Scholar,9Huang Y.J. Nelson C.E. Brodie E.L. et al.Airway microbiota and bronchial hyperresponsiveness in patients with suboptimally controlled asthma.J Allergy Clin Immunol. 2011; 127 (372-381.e1-3)Abstract Full Text Full Text PDF Scopus (465) Google Scholar Furthermore, EBC is a simple and less invasive method for obtaining samples from the lung. This sample collection allows for more readily available analysis of the lung microbiome. Our model using AI techniques revealed that Firmicutes and Proteobacteria were common important features of asthma at the phylum level. It also revealed the best performance for differentiating patients with asthma from healthy controls. However, this technique is more complex and costly compared than the usual criteria for diagnosing asthma. Therefore, this method may have limited helpfulness when conventional methods, such as spirometry and the bronchial provocation test, are not available for diagnosing asthma. The ability of this tool should be further verified to differentiate asthma from chronic obstructive pulmonary disease and nonasthmatic bronchitis. Despite these challenges, modulating the lung microbiome in patients with asthma has the capacity to act as a new treatment strategy.7Shukla S.D. Shastri M.D. Chong W.C. et al.Microbiome-focused asthma management strategies.Curr Opin Pharmacol. 2019; 46: 143-149Crossref PubMed Scopus (5) Google Scholar Our model may be used as an indicator of screening subjects with the potential of developing asthma because dysbiosis of the microbiome early in life is related to the development of asthma later in life. Several studies have mentioned the potential role of the microbiome in shaping phenotypes, endotypes, and disease severity of asthma with respect to patient-specific therapy.10Loverdos K. Bellos G. Kokolatou L. et al.Lung microbiome in asthma: current perspectives.J Clin Med. 2019; 8: 1967Crossref Scopus (17) Google Scholar However, our further subgroup analyses did not reveal significant differences according to neutrophil airway inflammation, severity and duration of asthma, or use of inhaled corticosteroids. This may be due to the small sample size with heterogeneous clinical characteristics. Providing the clinical use of microbiome-based diagnosis and management may require a longitudinal prospective study of sufficient size. Our study was limited in that dietary patterns, use of antibiotics, and recent lower respiratory infections were not evaluated as confounding factors. In addition, we did not present the specific mechanisms or functional activity underlying microbial influences on immunomodulatory effects. In conclusion, we have revealed differences between the lung microbiome of individuals with asthma and controls by analyzing EVs derived from EBC. We also proposed an asthma diagnostic model using AI. Although the clinical application of microbiota-targeting management is controversial, our findings may support a microbiome-based approach for diagnosing and managing asthma.