Genomic context of resistance determinants in 150 bacterial genomes across six taxa: derived data and evaluation evidence

What was done. 150 bacterial genomes were assembled de novo from public short-read data across six bacterial taxa (three species-level and three genus-level) (Klebsiella pneumoniae, Enterobacter spp., Citrobacter spp., Serratia spp., Enterococcus faecium and E. faecalis). Each isolate was paired with its own closed reference genome, and the 9,784 of 19,320 assembled contigs that met the 1 kb eligibility floor were aligned back to that reference to establish the replicon from which each derives. Resistance determinants were annotated on the same contigs, providing the genomic context for every determinant in the cohort. Those labels then served as the truth for a prospectively sealed, truth-blind evaluation of 12 third-party classifiers and predeclared baselines, with predictions frozen and self-re-executed twice by the same operator with no second participant before any truth artifact was introduced into the analysis system. Principal derived result. Of 635 resistance-gene-bearing contigs, 266 (41.9%) are plasmid-derived. The proportion ranges from 57.4% in Citrobacter spp. to 7.1% in Serratia spp., and across gene families, the distribution is bimodal: of the 40 families with at least 10 contigs, 15 are at or above 90% plasmid-derived, and 14 are at or below 10%. These are truth counts derived from alignment and are independent of any classifier. What this deposit does not contain. Raw sequencing reads are public archive data. They are cited by run, sample and assembly accession with the MD5 checksum of every file as verified at acquisition, permitting deterministic re-retrieval, and they are not redistributed here. The native, unparsed outputs of the classifier panel are retained by the author and are available on request; they are not required to reproduce any published value, for which the parsed calls in Part 1 suffice. Integrity. Every file carries a SHA-256 in MANIFEST.sha256. Parts 1 and 2 are deposited byte-for-byte as they were built on the execution host, so their digests match the values recorded at the results freeze rather than any values computed during the preparation of this deposit.