Independent evaluation of PlasmidCall in 432 carbapenem-resistant Enterobacterales isolates: native classifier outputs and execution records
Year: 2026
Type: dataset
Abstract: This dataset accompanies the manuscript "PlasmidCall: a plasmid–chromosome classification model for genomic context assignment of antimicrobial resistance determinants with independent evaluation in carbapenem-resistant Enterobacterales". It holds the native outputs and execution records of the classifier panel run in the independent evaluation of PlasmidCall under a time-stamped analysis plan: twelve plasmid classifiers and AMRFinderPlus, run on the short-read assemblies of the 432 isolates of the independent evaluation cohort (NCBI BioProject PRJNA1133668). The outputs were produced before any Nanopore read of the cohort was retrieved. With these records, every classifier call used in the evaluation, including the calls on database-absent contigs, can be traced to the output its tool wrote. Contents. The native output directory of each of the 5,616 runs (432 isolates, each run through the twelve classifiers and AMRFinderPlus), kept as the tool wrote it. For each run, a receipt that records the container image and its image ID, the full container command, the start and end time, the exit code and the status; a SHA256SUMS file with the SHA-256 of every output file; the command file; and the run log. The parsed call table of each classifier and isolate, in which every row carries the SHA-256 of its run's SHA256SUMS file. The panel runner as each machine ran it, with its SHA-256, and the state marker of every run. Eleven runs failed (six gplas2, three MOB-recon, two PlasmidFinder); their receipts record the failure, and no call was taken from them. Classifier versions and databases are listed in Table 1 of the manuscript; AMRFinderPlus was v4.2.7 (database 2026-08-07.1). The dataset does not contain the assemblies, reference labels, fixed predictions, evaluation results, development models or analysis code. Two sources. The run was split between a cloud server (383 isolates; files prefixed server_, tar + zstd) and a workstation (49 isolates; files prefixed workstation_, tar + gzip). Each source has its own README.txt and a MANIFEST.tsv that gives the byte size and SHA-256 of each archive as recorded on 2026-09-24. The server's README.txt and MANIFEST.tsv also list assemblies.tar.zst, which is not part of this dataset; the short-read assemblies are in the evaluation-data deposit. Records kept as written. The archives keep the records as the runs wrote them, including the absolute file paths of the machines that ran each step, the label "P1.13 truth-blind" that the runner carried over from the earlier 150-genome evaluation, and earlier internal names for the study. README.md explains how to check each run's SHA256SUMS file against an extracted archive. Split archives. Archives larger than 2 GiB are stored in parts named, for example, server_native_genomad.tar.zst.part01. Rejoin them in order, for example cat server_native_genomad.tar.zst.part* > server_native_genomad.tar.zst, then run sha256sum -c --ignore-missing REJOINED.sha256. MANIFEST.sha256 gives the SHA-256 of every other file, parts included. Related deposits. The evaluation data (short-read and hybrid assemblies, reference labels, fixed predictions, evaluation results and the time-stamped analysis plan): doi:10.6084/m9.figshare.34021401. PlasmidCall release v1.1.0, with the v1.2-General and v1.1 models: doi:10.6084/m9.figshare.34018380.v1. The four development models, with the code that applied them, are version 2 of the same item: doi:10.6084/m9.figshare.34018380.v2. The analysis code: doi:10.6084/m9.figshare.34063578. The data of the earlier 150-genome evaluation: doi:10.5281/zenodo.22086357.
Source: Harvard Dataverse
DOI: 10.7910/dvn/q5jnx9
Author: Piranfar, Vahhab
Language: English
Subfield: Microbiology (medical)
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7220358293