Machine Learning for Antimicrobial Resistance Prediction

Abstract

Antimicrobial resistance threatens modern medicine by rendering antibiotics ineffective against bacterial pathogens, and machine learning models that predict resistance phenotypes from bacterial genome sequences offer rapid alternatives to culture-based susceptibility testing, yet adoption into clinical microbiology varies across ML approaches and pathogen contexts. We evaluated 216 ML-based AMR prediction programmes across centres in France and Spain between 2018 and 2022, spanning five approach categories: rule-based resistance gene detection with ML scoring, whole-genome machine learning classifiers, deep learning sequence models, protein structure-informed resistance prediction, and multi-drug resistance co-occurrence networks. An AMR ML Prediction Index (AMPI) was constructed from five sub-scores -- susceptibility prediction accuracy, cross-pathogen generalisation, novel resistance mechanism detection, clinical turnaround time, and actionable reporting quality -- with weights from regression against sustained adoption into clinical microbiology pipelines. AMPI correlated with adoption at r = +0.84 and discriminated adopted from non-adopted methods with an AUC of 0.882. Rule-based detection with ML scoring scored highest (mean AMPI 0.824), while multi-drug co-occurrence networks trailed at 0.598. Only 35.5 percent exceeded the 0.75 threshold. Prediction accuracy carried the largest weight (beta = +0.278), followed by cross-pathogen generalisation (beta = +0.230).