BMC microbiology · 2026
Combining mass spectrometry and machine learning models for predicting Klebsiella pneumoniae antimicrobial resistance: a multicenter experience from clinical isolates in Italy.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Rocchi Ettore, Nicitra Emanuele, Calvo Maddalena, Cento Valeria, Peiretti Laura, Asif Zian, Menchinelli Giulia, Posteraro Brunella, Sala Claudia, Colosimo Claudia, Cricca Monica, Sambri Vittorio, Sanguinetti Maurizio, Castellani Gastone, Stefani Stefania
Аннотация
BACKGROUND: Multidrug-resistant Klebsiella pneumoniae represents a significant challenge in healthcare settings, prompting numerous studies on the rapid detection of antimicrobial resistance. Mass spectrometry has been recently integrated into routine laboratory diagnostics, providing highly sensitive results for pathogen identification. Furthermore, previously published studies have demonstrated its potential application in predicting antimicrobial resistance.
MATERIALS AND METHODS: The study collected 686 clinical isolates of K. pneumoniae from three Italian hospitals and used their MALDI-TOF mass spectra as input to machine learning models for predicting susceptibility profiles to amikacin and meropenem, which were selected as the most represented antibiotic molecules within the aminoglycoside and carbapenem classes, commonly used for the treatment of K. pneumoniae infections. After preprocessing, K. pneumoniae spectra were fed to machine learning classifiers within a nested cross-validation framework. Several performance metrics were computed to compare models and identify the most appropriate one for each antibiotic. Given the multicentric nature of the study, a batch-effect correction step was applied to reduce site-specific variability using the in-house developed Python package combatlearn (available on GitHub: https://github.com/EttoreRocchi/combatlearn ).
RESULTS: The XGBoost model achieved the best performance for both antibiotics (AUROC = 0.822 ± 0.028 for amikacin; AUROC = 0.887 ± 0.019 for meropenem). A per-site performance analysis revealed that, while performances' variability was inherently linked to each center's sample size, combatlearn-based harmonization effectively aligned mean AUROC values across sites.
CONCLUSIONS: Our study demonstrates the capability of MALDI-TOF mass spectra to predict amikacin and meropenem resistance in K. pneumoniae directly from clinical spectra, supporting its potential as a rapid and cost-effective approach for both antimicrobial resistance surveillance through machine learning models and clinical decision support in routine microbiology practice.
MATERIALS AND METHODS: The study collected 686 clinical isolates of K. pneumoniae from three Italian hospitals and used their MALDI-TOF mass spectra as input to machine learning models for predicting susceptibility profiles to amikacin and meropenem, which were selected as the most represented antibiotic molecules within the aminoglycoside and carbapenem classes, commonly used for the treatment of K. pneumoniae infections. After preprocessing, K. pneumoniae spectra were fed to machine learning classifiers within a nested cross-validation framework. Several performance metrics were computed to compare models and identify the most appropriate one for each antibiotic. Given the multicentric nature of the study, a batch-effect correction step was applied to reduce site-specific variability using the in-house developed Python package combatlearn (available on GitHub: https://github.com/EttoreRocchi/combatlearn ).
RESULTS: The XGBoost model achieved the best performance for both antibiotics (AUROC = 0.822 ± 0.028 for amikacin; AUROC = 0.887 ± 0.019 for meropenem). A per-site performance analysis revealed that, while performances' variability was inherently linked to each center's sample size, combatlearn-based harmonization effectively aligned mean AUROC values across sites.
CONCLUSIONS: Our study demonstrates the capability of MALDI-TOF mass spectra to predict amikacin and meropenem resistance in K. pneumoniae directly from clinical spectra, supporting its potential as a rapid and cost-effective approach for both antimicrobial resistance surveillance through machine learning models and clinical decision support in routine microbiology practice.