Biochip-simulated genotype signals enable accurate and interpretable AMR prediction via machine learning

Area of Science:

• Computational biology and bioinformatics
• Machine learning applications in healthcare
• Genomic data analysis for infectious diseases

Background:

• Antimicrobial resistance (AMR) is a critical global health threat.
• Traditional AMR diagnostics are slow and require specialized equipment.
• Need for rapid, accurate, and interpretable diagnostic tools.

Purpose of the Study:

• Develop a smart pathogen sensing framework using simulated biochip genotypic signals.
• Integrate machine learning (ML) and explainable AI (XAI) for AMR prediction.
• Enable model interpretability and personalized feedback for clinical decision support.

Main Methods:

• Generated synthetic biochip-like signals from 10,000 Salmonella enterica AMR genotype profiles.
• Employed KMeans clustering for unsupervised subtype discovery.
• Trained supervised models (Random Forest, XGBoost, Voting Classifier) with cross-validation and utilized SHAP values for explainability.

Main Results:

• The Voting Classifier demonstrated superior multi-class prediction performance (accuracy, precision, recall, F1-score, AUC).
• UMAP and silhouette scores confirmed robust clustering of pathogen subtypes.
• SHAP interpretation identified key resistance genes, and a rule-based system provided actionable clinical insights.

Conclusions:

• Presents a scalable, proof-of-concept diagnostic framework for AMR.
• Integrates simulated biochip data, interpretable ML, and a recommendation system.
• Offers a pathway toward clinically relevant AMR diagnostics with enhanced decision support.