Machine learning for exploring microbial inter-kingdom associations in Cystic Fibrosis and Bronchiectasis

Abstract

BackgroundCystic fibrosis (CF) and non-CF bronchiectasis (BX) are characterised by severe chronic infections. Fungal and bacterial components of infection are both recognized. Little however is known about how fungal and bacterial organisms interact and whether these interactions impact on disease outcomes. MethodsQuantitative PCR and next-generation sequencing of ITS2 and 16S rRNA gene was carried out on 107 patients with CF or BX with clinically defined fungal infection status for all patients. The relationship between fungal and bacterial community composition was extensively explored using: random forest modelling, correlation network analysis, multi-omics factor analysis, and sample-wise clustering, to understand associations both within and between the microbial communities and their relationship to respiratory disease. ResultsRandom forest modelling demonstrated distinct fungal and bacterial communities within CF and BX patients. The inclusion of both kingdoms in the models did not improve discrimination between the two diseases. Within the CF patients, bacterial community composition was independent of clinical fungal disease status. Bacterial and fungal communities did not relate to the presence of CF pulmonary exacerbations (CFPE). Correlation network analysis found intra-kingdom interactions were predominant in the data. Multi-omics factor analysis (MOFA) revealed latent factors corresponding to single kingdoms. Thus, in the bacterial community we identified two distinct clusters characterised by the presence or absence of Pseudomonas-domination. This was independent of fungal community which was characterised by a second set of independent clusters dominated by Saccharomycetes. ConclusionsIn this study we were unable to detect clear evidence of clinically significant inter-kingdom interactions between the bacterial and fungal communities. While further work is required to fully understand microbial interaction within the lung, our data suggests that interkingdom interactions may not be the primary driver of patient outcomes, particularly in the context of fungal infection.