Metabolic Network-Guided Binning of Metagenomic Sequence Fragments

There is a pressing need for methods that can reliably reconstruct species-level genomes from metagenomic samples to address questions in ecology, bioremediation, and infectious disease and human health. We present the Sorting by Network Completion (SoNeC) approach for assigning reactions to incomplete metabolic networks based on a gap fill score. We successfully demonstrate proof of concept in a set of 100 genome-scale metabolic network reconstructions, and delineate the variables that impact the reaction assignment accuracy, including the size of the constituent genome pool, the initial completeness of the genome bins, the reaction group size during assignment, and the fraction of genomes that are not considered as bins. We demonstrate the integration of SoNeC with existing approaches (such as cross-sample contig abundance profiles) on a set of 94 metagenomic samples from the Human Microbiome Project. We show that not only does SoNeC aid in reconstructing species-level genomes, but also improves the functional predictions that can be made with the resulting metabolic network reconstructions. This work delineates an independent source of information that allows more complete reconstruction of genomes from metagenomic sequence data, thus illuminating the inner workings of organisms and communities with ecological, medical, and industrial importance. The datasets and code are available at https://bitbucket.org/mattbiggs/sorting_by_network_completion/wiki/Home.