Rapid Genomic Prediction of Antibiotic Resistance in Klebsiella pneumoniae Using Al-Rayan Deep Learning model: an Artificial Intelligence Approach

Abstract

Background: Antimicrobial resistance in Klebsiella pneumoniae causes prolonged hospital stays and increased mortality. Current phenotypic testing requires 48-72 hours, delaying appropriate antibiotic therapy.

Objective: To develop a deep learning model (Al-Rayan Deep Learning Model) for rapid prediction of antibiotic resistance from genomic data, enabling same-day targeted therapy.

Methods: We analyzed 141,718 K. pneumoniae clinical isolates using a novel deep learning framework. The model processes genomic data through optimized feature selection and group-aware validation to prevent data leakage. Performance was evaluated on an independent test set of 25,718 isolates from 23,548 unique patient groups.

Results: The model achieved exceptional performance with AUC-ROC of 0.990 and average precision of 0.999. For resistant isolates, it demonstrated perfect precision (1.00) and high recall (0.94), correctly identifying all truly resistant cases while minimizing false positives. The framework identified 50 key resistance genes driving predictions, providing biological plausibility.

Conclusion: This deep learning approach enables accurate, rapid resistance prediction within hours using genomic sequencing data. While current sequencing costs limit widespread use to critical care settings, the technology offers significant potential for antibiotic stewardship programs.