On the estimation of metabolic profiles in metagenomics

Metagenomics enables the characterization of the specific metabolic potential of a microbial community. The common approach towards a quantitative representation of this potential is to count the number of metagenomic sequence fragments that can be assigned to metabolic pathways by means of predicted gene functions. The resulting pathway abundances make up the metabolic profile of the metagenome and several different schemes for computing these profiles have been used. So far, none of the existing approaches actually estimates the proportion of sequences that can be assigned to a particular pathway. In most publications of metagenomic studies, the utilized abundance scores lack a clear statistical meaning and usually cannot be compared across different studies. Here, we introduce a mixture model-based approach to the estimation of pathway abundances that provides a basis for statistical interpretation and fast computation of metabolic profiles. Using the KEGG database our results on a large-scale analysis of data from the Human Microbiome Project show a good representation of metabolic differences between different body sites. Further, the results indicate that our mixture model even provides a better representation than the dedicated HUMAnN tool which has been developed for metabolic analysis of human microbiome data.