Electronic supplementary material
The online version of this article (https://doi.org/10.1186/s12967-017-1365-7) contains supplementary material, which is available to authorized users.
Anirban Sinha and Koundinya Desiraju contributed equally to this work
Asthma is a complex, heterogeneous disorder with similar presenting symptoms but with varying underlying pathologies. Exhaled breath condensate (EBC) is a relatively unexplored matrix which reflects the signatures of respiratory epithelium, but is difficult to normalize for dilution.
Here we explored whether internally normalized global NMR spectrum patterns, combined with machine learning, could be useful for diagnostics or endotype discovery. Nuclear magnetic resonance (NMR) spectroscopy of EBC was performed in 89 asthmatic subjects from a prospective cohort and 20 healthy controls. A random forest classifier was built to differentiate between asthmatics and healthy controls. Clustering of the spectra was done using k-means to identify potential endotypes.
NMR spectra of the EBC could differentiate between asthmatics and healthy controls with 80% sensitivity and 75% specificity. Unsupervised clustering within the asthma group resulted in three clusters (n = 41,11, and 9). Cluster 1 patients had lower long-term exacerbation scores, when compared with other two clusters. Cluster 3 patients had lower blood eosinophils and higher neutrophils, when compared with other two clusters with a strong family history of asthma.
Asthma clusters derived from NMR spectra of EBC show important clinical and chemical differences, suggesting this as a useful tool in asthma endotype-discovery.