Exploring bacterial diversity via a curated and searchable snapshot of archived DNA sequences

The open sharing of genomic data provides an incredibly rich resource for the study of bacterial evolution and function and even anthropogenic activities such as the widespread use of antimicrobials. However, these data consist of genomes assembled with different tools and levels of quality checking, and of large volumes of completely unprocessed raw sequence data. In both cases, considerable computational effort is required before biological questions can be addressed. Here, we assembled and characterised 661,405 bacterial genomes retrieved from the European Nucleotide Archive (ENA) in November of 2018 using a uniform standardised approach. Of these, 311,006 did not previously have an assembly. We produced a searchable COmpact Bit-sliced Signature (COBS) index, facilitating the easy interrogation of the entire dataset for a specific sequence (e.g., gene, mutation, or plasmid). Additional MinHash and pp-sketch indices support genome-wide comparisons and estimations of genomic distance. Combined, this resource will allow data to be easily subset and searched, phylogenetic relationships between genomes to be quickly elucidated, and hypotheses rapidly generated and tested. We believe that this combination of uniform processing and variety of search/filter functionalities will make this a resource of very wide utility. In terms of diversity within the data, a breakdown of the 639,981 high-quality genomes emphasised the uneven species composition of the ENA/public databases, with just 20 of the total 2,336 species making up 90% of the genomes. The overrepresented species tend to be acute/common human pathogens, aligning with research priorities at different levels from individual interests to funding bodies and national and global public health agencies.

This study presents the first uniformly assembled, comprehensively described and searchable dataset of 661,405 bacterial genomes; this resource will empower more scientists to harness the multitude of data in public sequencing archives, but also reveals the biased composition of these archives, with 90% of the data originating from just 20 species.