Evaluating a Rule-Based Gut Resistome-to-Antibiotic Guidance Framework: Performance, Failure Modes and a Revised Decision Rule

A companion paper specified the architecture of a six-stage computational framework that converts gut-bacterial resistome content into class-level antibiotic guidance. Here we report the quantitative behaviour of that framework when executed over a reference panel of ten gut-relevant bacterial strains and six curated resistance determinants. Seven determinant-present calls were made across the panel (mean burden 0.7 determinants per strain), generating 96 guidance decisions across sixteen clinical antimicrobial classes. Under an original binary decision rule, seven decisions correctly recommended avoidance and eighty-nine asserted that an agent was a preferred candidate; systematic checking against established clinical microbiology showed several of the latter to be false, including a recommendation of cephalosporins for an intrinsically resistant enterococcus and aminoglycosides for obligate anaerobes. A revised three-state rule, supported by an independently compiled intrinsic resistance register, removed every identified error and reduced the number of positive-suitability claims from eighty-nine to zero. Standard descriptive measures of resistome composition — Shannon diversity, ordination, co-occurrence networks, association testing and classifier accuracy — were computed and shown to be uninformative or statistically inapplicable at this panel size, a finding we report as a methodological result rather than a limitation to be glossed over. A forty-community synthetic simulation confirmed that the underlying burden and diversity measures respond in the expected direction to specified compositional differences, while also exposing an internal inconsistency in the permutational test used to compare community dissimilarity. We conclude that resistome profiling of this kind can support avoidance decisions but cannot, on present evidence, support any positive claim of suitability, and we set out the conditions that would need to be satisfied before that could change.