A Bayesian compositional estimator for microbial taxonomy based on biomarkers

Abstract

Determination of microbial taxonomy based on lipid or pigment spectra requires use of a compositional estimator. We present a new approach based on Bayesian inference and an implementation in the open software platform R. The Bayesian Compositional Estimator (BCE) aims not only to obtain a maximum likelihood solution, but also to provide a complete estimate of the taxonomic composition, including probability distributions and dependencies between estimated values. BCE results are compared with those obtained with CHEMTAX. The BCE has not only a similar accuracy, but also extracts more information from the data, the most obvious being standard deviation and covariance estimates