A data-quality audit of adeIJK calls in public Acinetobacter baumannii genome-MIC data

Binary genotype-minimum inhibitory concentration (MIC) regression requires an observable presence-absence contrast. We audited source identity, uniform annotation, locus completeness and binary variation in 833 public records labelled as Acinetobacter baumannii. Fresh RGI/CARD annotation called the adeIJK composite, requiring adeI, adeJ and adeK, in 831/833 records (99.76%), including all 826 reference-consistent records. Two earlier empty RGI outputs explained the change from 829/833 in the submitted analysis. Locus review classified 816 eligible records as PRESENT and 10 as UNCERTAIN, with no confirmed ABSENT records in any antibiotic subset. The adeIJK-specific interval-censored coefficient was not estimable, so no association test was performed. Core-SNP modelling changed variance allocation across MIC models, while Mash-kernel repair affected conditioning. This single-dataset audit illustrates the need to preserve source identifiers, verify annotation outputs and resolve uncertain loci before binary genotype-MIC modelling.