Machine Learning for Antimicrobial Resistance Prediction: Current Practice, Limitations, and Clinical Perspective.

Area of Science:

• Genomics
• Computational Biology
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) is a critical global health threat requiring novel prevention strategies.
• Advancements in pathogen genome sequencing provide large datasets for analysis.
• Machine learning (ML) shows promise in predicting AMR based on genomic data.

Purpose of the Study:

• To advocate for integrating ML into frontline settings for AMR prediction.
• To identify necessary refinements for safe and confident ML implementation.
• To address limitations hindering the clinical adoption of ML for AMR.

Main Methods:

• Utilizing large pathogen genome datasets for ML model training.
• Analyzing gene content and genome composition as predictors of AMR.
• Evaluating ML model performance and limitations.

Main Results:

• ML models can predict AMR using genomic information.
• Current ML models have limitations, including treating genes independently and potential inaccuracies with new mutations.
• Transparency and explainability are crucial for end-user trust.

Conclusions:

• ML holds significant potential for predicting AMR and aiding in diagnostic implementation.
• Further research must focus on refining ML models for accuracy, robustness, and interpretability.
• Bridging the gap between ML development and diagnostic application requires addressing current limitations.