Machine learning-based prediction of multi-level antimicrobial resistance in Klebsiella pneumoniae using whole-genome sequencing data

This study presents an advanced machine learning approach (MLWA) that uses whole-genome sequencing (WGS) data from over 5,200 bacterial strains to accurately predict antimicrobial resistance across multiple levels, including resistant/susceptible, intermediate categories, and resistance intensity. The models demonstrated high performance (AUC > 0.9, ~96% agreement) across 11 antibiotics and remained robust across different regions, strain types, and time periods. With low error rates and strong generalizability, this approach significantly improves genotype-to-phenotype prediction and enables faster, more precise antibiotic selection, highlighting its strong potential for clinical application in AMR management.