PlasmidCall: predicting whether resistance genes in short-read assemblies lie on plasmids or chromosomes, tested in 432 carbapenem-resistant Enterobacterales
Year: 2026
Type: preprint
Abstract: Antimicrobial resistance genes (ARGs) can lie on the bacterial chromosome, where they are passed on mainly with the host strain, or on plasmids, which can move them between bacteria. Most bacterial genomes are sequenced with short reads, which assemble into many fragments (contigs), so whether a contig, and any ARG on it, comes from a plasmid or from the chromosome usually has to be predicted. Plasmid classifiers are software tools that make this prediction for each contig from different kinds of evidence, and each tends to fail on different contigs. PlasmidCall v1.2-General is a logistic-regression model that combines the plasmid or chromosome calls of twelve such classifiers into one decision. Four gradient-boosting models, fitted on a larger training set, combine some or all of these calls with features of the contig, its neighbours in the assembly graph and its isolate. The five models were tested without refitting on 432 carbapenem-resistant Enterobacterales isolates from one hospital, none used in model development. For each isolate, a hybrid assembly built from long and short reads provided reference labels (plasmid or chromosome) for 30,986 of its 35,805 short-read contigs of at least 1 kb, whether or not they carried an ARG. PlasmidCall v1.2-General gave a decision on every contig: 94.9% of its plasmid calls were correct (precision 0.9491; 95% CI 0.8915-0.9642, resampling clusters of closely related isolates), and it found 91.8% of plasmid contigs (recall 0.9184). Its precision was 0.028 lower than a previously reported estimate; the two 95% intervals overlapped when clusters of related isolates were resampled, but not when single isolates were. The gradient-boosting models found 98.9-99.2% of plasmid contigs at precision 0.887-0.922. On contigs carrying ARGs, v1.2-General had precision 0.935 and recall 0.969, and on the 383 contigs carrying the carbapenemase gene blaKPC (post hoc), 0.994 and 0.970. PlasmidCall can therefore place resistance genes, including blaKPC, on a plasmid or the chromosome from short-read data alone, with a choice between higher precision and higher recall.
Source: Research Square
Author: Vahhab Piranfar
Language: English
Subfield: Microbiology (medical)
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7225735501