CarbaDetector: a machine learning model for detecting carbapenemase-producing Enterobacterales from disk diffusion...

Area of Science:

• Medical Microbiology
• Computational Biology
• Infectious Diseases

Background:

• Carbapenemase-producing Enterobacterales (CPE) represent a critical global health threat.
• Current methods for CPE detection are often slow and challenging.
• Accurate and rapid detection is crucial for effective infection control and treatment.

Purpose of the Study:

• To develop and validate a machine learning model (CarbaDetector) for predicting carbapenemase production in Enterobacterales.
• To compare the performance of CarbaDetector against existing standard algorithms.
• To assess the potential of CarbaDetector in reducing confirmatory testing and accelerating time to result.

Main Methods:

• Development of a random-forest machine learning model (CarbaDetector).
• Training dataset comprised 385 isolates with whole genome sequencing as reference.
• Validation on two independent external datasets (282 and 518 isolates).
• Prediction based on inhibition zone diameters of eight antibiotics.

Main Results:

• CarbaDetector demonstrated high performance: 96.6% sensitivity and 84.4% specificity on the training set.
• External validation showed excellent performance: 96.3% sensitivity/86.1% specificity (Dataset A) and 91.2% sensitivity/87.0% specificity (Dataset B).
• CarbaDetector significantly outperformed EUCAST and CA-SFM algorithms in specificity (8.2% and 40.1% respectively on training data).

Conclusions:

• CarbaDetector offers a sensitive and specific method for predicting carbapenemase production.
• The model, available as a web-app, can reduce unnecessary confirmatory tests and speed up results.
• CarbaDetector has the potential to enhance CPE detection, particularly in resource-limited settings.