Towards an Interpretable Machine Learning Model for Predicting Antimicrobial Resistance

This paper explores how interpretable machine learning (ML) can improve the prediction of antimicrobial resistance (AMR) by integrating phenotype–genotype synergy. While ML methods already achieve high accuracy in predicting resistance from genomic data, most overlook complex, nonlinear interactions between genes and mutations. By accounting for these synergistic effects, models can better capture biological realities, improve transparency, and enhance clinical trust.

The authors highlight three pillars of interpretability:

• assessing individual genetic or phenotypic features,

• ensuring traceability of predictions, and

• understanding feature interactions.

They review existing ML approaches for AMR prediction, discuss challenges such as limited datasets, overfitting, and biological variability, and propose workflows that combine computational predictions with experimental validation.

Conclusion: Interpretable ML models that integrate genotype–phenotype interactions are key to more reliable AMR prediction and could accelerate the discovery of effective treatments against resistant pathogens.