Benchmarking of shotgun sequencing depth highlights limitations of strain-level analysis and shallow metagenomics

Abstract Shallow metagenomics promises taxonomic and functional insights into samples at an affordable price. To determine the depth of sequencing required for specific analysis, benchmarking is required using defined microbial communities. We used complex mixtures of DNA from cultured gut bacteria and analysed taxonomic composition, strain-level resolution, and functional profiles at up to eleven sequencing depths (0.1-50.0 Gb). Reference-based analysis provided accurate taxonomic, and strain-level insights at 0.5-1.0 Gb. In contrast, de-novo metagenome-assembled genome (MAG) reconstruction required deep sequencing (>10 Gb), and even high-quality MAGs were chimeric, with 54.5 to 81.8 % accurately representing the original strains, depending on the bioinformatic approach used. However, the issue of chimeric MAGs can be reduced by using strain-aware assembly methods or long-read sequencing. Functionally, 2 Gb provided reliable insights at the pathway level, but sufficient proteome coverage was only achieved at or above 10 Gb. Library preparation and host DNA contamination were identified as confounders in shallow metagenomic analysis. This comprehensive analysis using complex mock communities provides guidance to an increasing community of scientists interested in using shallow metagenomics, and highlights the limitations of MAGs in accurately capturing strain-level diversity. Competing Interest Statement The authors have declared no competing interest. Footnotes Additional data and analyses have been included: 1. Analysis of a new, complex mock community of 70 species in which the abundances were staggered, was added. 2. Two additional sequencing depths (20 and 50 Gb) were added to assess MAG reconstruction also after deeper sequencing. 3. In-silico subsampling of the 50 Gb sample in multiple replicates was performed, followed by statistical testing. 4. Additional to the assembly tool MEGAHIT, metaSPAdes was used. 5. Effect of multi-coverage binning on MAG constuction was tested. 6. Effect of long-read sequencing on MAG constuction was tested. 7. Unsupervised taxonomic profiling by MetaPhlAn was added. https://github.com/ClavelLab/Benchmarking-shallow-Metagenomics