Bayesian estimation of bacterial community composition from 454 sequencing data

Estimating bacterial community composition from a mixed sample in different applied contexts is an important task for many microbiologists. The bacterial community composition is commonly estimated by clustering polymerase chain reaction amplified 16S rRNA gene sequences. Current taxonomy-independent clustering methods for analyzing these sequences, such as UCLUST, ESPRIT-Tree and CROP, have two limitations: (i) expert knowledge is needed, i.e. a difference cutoff between species needs to be specified; (ii) closely related species cannot be separated. The first limitation imposes a burden on the user, since considerable effort is needed to select appropriate parameters, whereas the second limitation leads to an inaccurate description of the underlying bacterial community composition. We propose a probabilistic model-based method to estimate bacterial community composition which tackles these limitations. Our method requires very little expert knowledge, where only the possible maximum number of clusters needs to be specified. Also our method demonstrates its ability to separate closely related species in two experiments, in spite of sequencing errors and individual variations.