Comparative assessment of annotation tools reveals critical antimicrobial resistance knowledge gaps in Klebsiella pneumoniae

This study  shows that current databases and annotation pipelines used to identify antibiotic-resistance markers vary widely in completeness and predictive power. By applying so-called “minimal models” (machine-learning models built only on known resistance determinants) to 20 major antimicrobials in K. pneumoniae, the authors found that for many antibiotics the known markers still fail to reliably predict resistance. This highlights both (a) the need for discovering novel resistance markers and (b) the importance of establishing standardised benchmarking datasets and pipelines in AMR-genomics for more accurate prediction, annotation-tool evaluation and future algorithm development.