show Abstracthide AbstractThe assembly of high-quality genomes from mixed microbial samples is a long-standing problem in genomics and metagenomics. Here, we describe the application of ProxiMetaTM, a Hi-C-based metagenomic deconvolution method, to deconvolve a human fecal sample. This method uses the intra-cellular proximity signal captured by Hi-C as a direct indicator of which sequences originated in the same cell, enabling culture-free de novo deconvolution of mixed genomes without any reliance on a priori information. We show that ProxiMeta provides incredibly high accuracy and sensitivity deconvolution, yielding 50 near-complete microbial genomes (many of which are novel) from a single fecal sample, and outperforms state-of-the-art contig binning approaches at high-quality genome reconstruction. ProxiMeta shows particularly good performance in constructing high-quality genomes for diverse but poorly-characterized members of the human gut, and offers additional information about microbial genome biology. We further use ProxiMeta to reconstruct genome plasmid content and gene sharing of plasmids – tasks that traditional binning methods often fail to accomplish. Our findings suggest that ProxiMeta-based deconvolution can be useful to a variety of applications in genomics and metagenomics.