A chemometric study on human breath mass spectra for biomarker identification in cystic fibrosis.

Alveolar breath samples from a small case-control study population have been collected and measured via ion-molecule reaction mass spectrometry, and a constructive statistical approach to the identification of volatile biomarkers has been formulated by applying multivariate statistical methods on the mass spectra. The nature of the data is such that the number of variables largely exceeds the observations, representing a typical experimental scenario when breath analysis is conducted using mass spectrometry. Principal components analysis has been performed on the high dimensional dataset of molecular abundances, providing evidence of case separation and reducing the number of functional discriminators by almost 90%. Afterwards, a deductive approach based on a binary regression was conducted on the reduced dataset, providing an entirely reliable case discrimination model exclusively depending on the concentrations in the breath mixture of 3 out of a total of 97 metabolites.