2025/05/14 by Axel Hamprecht, Linea Muhsal, Cansu Cimen van Dijk +7 · 1 voice
Biochemistry, Genetics and Molecular Biology · #Antibiotic Resistance in Bacteria #Bacterial Identification and Susceptibility Testing #Gut microbiota and health
paper · pdf · doi:10.21203/rs.3.rs-6630856/v1
crossref issued 2025/05/14 · crossref published 2025/05/14 · openalex publication_date 2025/05/14 · crossref created 2025/05/14 · openalex created_date 2025/10/10 · crossref deposited 2025/11/15 · crossref indexed 2026/04/23 · openalex updated_date 2026/07/14
Abstract Carbapenemase-producing Enterobacterales (CPE) are considered as the highest threat to global health in the WHO’s 2024 Bacterial Priority Pathogens List. Their detection is difficult, time-consuming and expensive. We developed a machine learning (ML) model using random-forest algorithms to predict carbapenemase-production based on inhibition zone diameters of eight antibiotics, utilizing 385 isolates for training. Whole genome sequencing served as reference standard. The model was validated on two external datasets (A = 282 isolates, B = 518 isolates). Sensitivity and specificity of the model were 97.5% and 84.4% on the primary dataset, 96.3% and 83.7% on dataset A, and 90.7% and 83.1% on dataset B (containing five antibiotics only). Comparatively, algorithms of the Antibiogram Committee of the French Society of Microbiology (CA-SFM) and of European Committee on Antimicrobial Susceptibility Testing (EUCAST) exhibited lower specificities, e.g. of 40.1% and 8.2% on the training dataset, respectively. The final model CarbaDetector can be accessed by a web app (https://uol.de/carba-detector), allowing users to input inhibition zone data for CPE prediction. The ML model offers high sensitivity and improved specificity, reducing unnecessary confirmatory testing which is helpful to accelerate the time to result and may improve diagnostics of CPE especially in resource limited settings.