Bayesian mixture analysis for metagenomic community profiling.

Abstract

Deep sequencing of clinical samples is now an established tool for the detection of infectious pathogens, with direct medical applications. The large amount of data generated provides an opportunity to detect species even at very low levels, provided that computational tools can effectively interpret potentially complex metagenomic mixtures. Data interpretation is complicated by the fact that short sequencing reads can match multiple organisms and by the lack of completeness of existing databases, in particular for viral pathogens. This interpretation problem can be formulated statistically as a mixture model, where the species of origin of each read is missing, but the complete knowledge of all species present in the mixture helps with the individual reads assignment. Several analytical tools have been proposed to approximately solve this computational problem. Here, we show that the use of parallel Monte Carlo Markov chains (MCMC) for the exploration of the species space enables the identification of the set of species most likely to contribute to the mixture. The added accuracy comes at a cost of increased computation time. Our approach is useful for solving complex mixtures involving several related species. We designed our method specifically for the analysis of deep transcriptome sequencing datasets and with a particular focus on viral pathogen detection, but the principles are applicable more generally to all types of metagenomics mixtures. The work is implemented as a user friendly R package, available from CRAN: http://cran.r-project.org/web/packages/metaMix.