Machine learning powered profiling: Rapid identification of Klebsiella Pneumoniae drug resistance from MALDI-TOF MS

Xiaobo Xu, Yuntao Gao,
Machine learning powered profiling: Rapid identification of Klebsiella Pneumoniae drug resistance from MALDI-TOF MS,
Journal of Microbiological Methods,
Volume 238,
2025,
107291,
ISSN 0167-7012,
https://doi.org/10.1016/j.mimet.2025.107291.
(https://www.sciencedirect.com/science/article/pii/S0167701225002076)
Abstract: Abstracts
The early identification of drug-resistant phenotypes in Klebsiella pneumoniae is essential for effective clinical intervention, infection management, and the prevention of resistance development and spread. This study aimed to construct multiple machine-learning models to rapidly and comprehensively predict susceptibility to nine common antimicrobial drugs using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) data of K. pneumoniae. A total of 484 K. pneumoniae isolates from Zhejiang Rongjun Hospital were collected and tested for in vitro susceptibility to nine antibiotics. Six supervised learning models were developed and evaluated: Random Forest, eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), Logistic Regression, Multilayer Perceptron, and Support Vector Machine. Performance was assessed using test score, ten-fold cross-validation, accuracy, specificity, F1 score, AUROC, and 95 %CI. The models performed best for Amikacin and Co-trimoxazole, whilst exhibiting the poorest predictive efficacy for levofloxacin. AdaBoost and XGBoost achieved high predictive performance with AUROC values ≥0.8 for all nine antimicrobial drugs. The XGBoost model demonstrated strong performance and stability across evaluation metrics. SHAP analysis based on the XGBoost model identified key features such as 4517.5 ± 2.5 Da for Amikacin and 6022.5 ± 2.5 Da for Co-trimoxazole. The study concluded that analyzing MALDI-TOF MS data with machine-learning models can rapidly predict the antibiotic susceptibility of K. pneumoniae, reducing resistance detection time to 1–2 h and accelerating clinical decision-making.
Keywords: Machine learning; Antimicrobial resistance; MALDI-TOF MS; Klebsiella pneumoniae