Bayesian mixture analysis for metagenomic community profiling.

Sofia Morfopoulou; Vincent Plagnol

Here, we present metaMix, a sensitive method for metagenomic species identification and abundance estimation. The method is implemented in an R package (http://cran.r-project.org/web/packages/metaMix). Using a Bayesian mixture model framework, we account for model uncertainty by performing model averaging and we resolve ambiguous assignments by considering all reads simultaneously. A key feature of the method is that it provides probabilities that answer pertinent biological questions, in particular the posterior probability for the presence of a species in the mixture. Additionally it accurately quantifies the relative proportions of the organisms. The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It . https://doi.org/10.1101/007476 doi: bioRxiv preprint This general framework is designed to address interpretation issues associated with closely related strains in the sample, low abundance organisms and absence of genomes from the reference database. We show that metaMix outperforms other methods in the community profiling task, particularly when complex structures with closely related strains are studied. As a consequence, it also produces more accurate relative abundance estimates for the species in the mixture. The method can deal with either unassembled reads or assembled contigs or both, allowing for flexibility of choice for the bioinformatics preprocessing. In practice, the choice of bioinformatics processing prior to the application of our Bayesian mixture analysis must be optimized for each application, and our processing pipeline has been designed with viral sequence identification from transcriptome sequencing as a main goal. Nevertheless, as demonstrated by our analysis of the mock bacterial community dataset, the method can be applied in other contexts.