Skip to main content
Download PDF

Rapid detection of carbapenem-resistant Escherichia coli and carbapenem-resistant Klebsiella pneumoniae in positive blood cultures via MALDI-TOF MS and tree-based machine learning models

BMC Microbiology volume 25, Article number: 44 (2025) Cite this article

Abstract

Background

Bloodstream infection (BSI) is a systemic infection that predisposes individuals to sepsis and multiple organ dysfunction syndrome. Early identification of infectious agents and determination of drug-resistant phenotypes can help patients with BSI receive timely, effective, and targeted treatment and improve their survival. This study was based on matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS), Decision Tree (DT), Random Forest (RF), Gradient Boosting Machine (GBM), eXtreme Gradient Boosting (XGBoost), and Extremely Randomized Trees (ERT) models were constructed to classify carbapenem-resistant Escherichia coli (CREC) and carbapenem-resistant Klebsiella pneumoniae (CRKP). Bacterial species were identified by MALDI-TOF MS in positive blood cultures isolated via the serum isolation gel method, and E. coli and K. pneumoniae in positive blood cultures were collected and placed into machine learning models to predict susceptibility to carbapenems. The aim of this study was to provide rapid detection of CREC and CRKP in blood cultures, to shorten the turnaround time for laboratory reporting, and to provide a basis for early clinical intervention and rational use of antibiotics.

Results

The collected MALDI-TOF MS data of 640 E. coli and 444 K. pneumoniae were analysed by machine learning algorithms. The area under the receiver operating characteristic curve (AUROC) for the diagnosis of E. coli susceptibility to carbapenems by the DT, RF, GBM, XGBoost, and ERT models were 0.95, 1.00, 0.99, 0.99, and 1.00, respectively, and the accuracy in predicting 149 E. coli-positive blood cultures were 0.89, 0.92, 0.90, 0.92, and 0.86, respectively. The AUROC for the diagnosis of K. pneumoniae susceptibility to carbapenems by the DT, RF, GBM, XGBoost, and ERT models were 0.78, 0.95, 0.93, 0.90, and 0.95, respectively, and the accuracy in predicting 127 K. pneumoniae-positive blood cultures were 0.76, 0.86, 0.81, 0.80, and 0.76, respectively.

Conclusions

Machine learning models constructed by MALDI-TOF MS were able to directly predict the susceptibility of E. coli and K. pneumoniae in positive blood cultures to carbapenems. This rapid identification of CREC and CRKP reduces detection time and contributes to early warning and response to potential antibiotic resistance problems in the clinic.

Clinical trial number

Not applicable.

Peer Review reports

Background

Bloodstream infection (BSI) has become one of the major public health burdens globally, as morbidity and mortality rates increase every year, prolonging hospital stays and increasing costs for healthcare systems [1]. In Europe, it is estimated that there are approximately 1.2 million episodes of BSI per year, and the number of BSI-related deaths per year ranges from 157,750 to 2,763,181 [2]. BSI, a serious systemic infectious disease, predisposes individuals to sepsis and multiple organ dysfunction syndrome if not treated promptly and appropriately [3]. Blood culture can help identify the pathogens responsible for bloodstream infections and is the gold standard for diagnosing bloodstream infections [4]. Rapid detection of pathogens in blood and identification of their drug-resistant phenotypes are crucial in bloodstream infections. Epidemiological data show that E. coli and K. pneumoniae are the top two gram-negative organisms in bloodstream infections, accounting for more than 50% of pathogens [5]. Among them, bloodstream infections caused by carbapenem-resistant Enterobacteriaceae (CRE) are on the rise and have become a serious threat to patients’ lives [6]. In 2024, the World Health Organization designated CRE as a critical group pathogen, as outlined in the organization’s priority bacterial pathogen list [7]. Both undertreatment (failure to cover the pathogen) and over-treatment (targeting but not isolating the resistant pathogen) against CRE result in increased clinical mortality rates [8]. Early identification of the infecting pathogen and further determination of the carbapenem phenotype can help patients with BSIs receive timely, effective and targeted treatment and improve patient survival rates [9].

Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) has revolutionized clinical diagnostics because of its ability to identify species quickly, reliably and cost-effectively and its potential to simplify microbial antimicrobial susceptibility testing (AST) [10, 11]. MALDI-TOF MS outperforms traditional diagnostic methods in terms of cost, speed and accuracy of microbial species identification, enabling bacterial identification without strain-passaging culture. Compared with traditional automated biochemical systems, the use of MALDI-TOF MS improves the workflow in clinical microbiology laboratories, reducing the time needed to obtain microbiological results by up to 24 h [12]. A machine learning algorithm is an algorithm that uses data and statistical techniques in a computer system to achieve automatic learning of patterns and regularities. Tree-based machine learning models are a class of predictive algorithms in machine learning that simulate the decision-making process by constructing a tree-like structure consisting of nodes and branches to classify or analyze data regressively. Each node in the model represents a decision rule for a feature, branches represent decision paths, and leaf nodes provide the final prediction [13].

Tree-based machine learning algorithms maximize the use of information contained in MALDI-TOF MS data to improve detection speed and accuracy through automated feature extraction and pattern recognition. The aim of this study was to achieve rapid detection of carbapenem-resistant Escherichia coli (CREC) and carbapenem-resistant Klebsiella pneumoniae (CRKP) in blood cultures by modelling and analysing MALDI-TOF MS data with tree-based machine learning models. This approach aims to shorten the turnaround time for laboratory reporting and provide a basis for early clinical intervention and rational use of antibiotics.

Methods

Strain collection and antimicrobial susceptibility testing

Nonduplicate strains of E. coli (n = 640) and K. pneumoniae (n = 444) tested at the Laboratory Department of Zhejiang Rong Jun Hospital (Jiaxing, China) between October 2021 and September 2024 were collected. E. coli and K. pneumoniae were cultured in Columbia blood Agar plates (Antu Biologicals, Zhengzhou, China) for 18–24 h. Bacterial identification was performed in MALDI-TOF MS (bioMérieux, Lyon, France). Antimicrobial susceptibility testing of bacteria with the VITEK®2 Compact Instrumentation (bioMérieux, Lyon, France) and the Gram-negative bacterial sensitization card AST-GN13 (bioMérieux, Lyon, France). Resistance to carbapenems were judged by the following criteria: MIC ≥ 2 µg/ml for ertapenem and MIC ≥ 4 µg/ml for imipenem or meropenem. The in vitro drug sensitivity judgment criteria of the strains were referred to CLSI M100 (33nd edition) [14].

MALDI-TOF MS identification

E. coli and K. pneumoniae were cultured on Columbia Blood Agar plates (Antu Biologicals, Zhengzhou, China) for 18–24 h. Single colonies of E. coli and K. pneumoniae were taken and applied to specific target plates, and 1 µl of α-cyano-4-hydroxycinnamic acid matrix solution was added and allowed to dry. Bacterial species identification was performed with a VITEK MS IVD system (bioMérieux, Lyon, France), which was subsequently matched against a database to determine the bacterial species. The number of shots reached 100, with m/z values ranging from 2000 Da to 20,000 Da, and peak plots with more than 90% identification were qualified. MALDI-TOF MS identification data were obtained once per bacterial strain.

Tree-based machine models

The mass spectral data of E. coli and K. pneumoniae collected by MALDI-TOF MS were processed. The mass spectral peaks were distributed in the region of 2000–20,000 Da, and the data were binned every 10 Da, and the maximum value of the peaks in the region was taken to represent the intensity of the peaks in the region. In vitro susceptibility to carbapenems was obtained according to the CLSI M100 judgement criteria, labelling resistance as “1” and susceptibility as “0” as outcome variables.

The mass spectrometry data of E. coli and K. pneumoniae were collected via MALDI-TOF MS, and the data were processed in jupyter on the basis of the python code imported into the corresponding modules of pyteomics, numpy, pandas, scipy, sklearn, matplotlib, imblearn and other libraries. When the original data “.mzML” format is converted to “.csv” format, Gaussian smoothing is performed on the data, which can reduce the “noise” in the data to a certain extent. The common points in model building include the introduction of the synthetic minority oversampling technique (SMOTE) algorithm to balance the differences between different sample sizes, the use of train_test_split to divide 70% of the dataset into the training set, and the division of 30% of the dataset into the test set. Setting random_state to 0 controls the randomness of the model. The roc_auc_score in sklearn.metrics is used to plot the area under the receiver operating characteristic curve (AUROC) and determine the optimal boundary point.

The evaluation metrics for the model include a test score on 30% of the original data, ten-fold cross validation score (an indicator of the model’s generalization ability), the AUROC of the model (an indicator of the overall assessment of the model), and accuracy scores of the model in positive blood culture data (external data to validate the model’s efficacy).

The tree-based models include Decision Tree (DT), Random Forest (RF), Gradient Boosting Machine (GBM), eXtreme Gradient Boosting (XGBoost), and Extremely Randomized Trees (ERT). The machine learning modelling process is shown in Fig. 1.

SHapley Additive explanation (SHAP) is a method for interpreting the prediction results of machine learning models. It is based on the concept of Shapley values in game theory, which assigns importance values to each feature of the model to explain the prediction process of the model. SHAP plots were drawn via TreeExplainer, which improved the interpretability of the model.

Fig. 1
figure 1

After obtaining the MALDI-TOF MS data, the 10 Da spectra were divided into one box, and the peaks within the same box were considered as a single property. The maximum peak intensity within each box was used as a representative value. Constructing Decision Tree, Random Forest, Gradient Boosting Machine, eXtreme Gradient Boosting, Extremely Randomized Trees Models

Full size image

Separation of pathogens by the serum separating gel method

The serum separating gel method was used for processing. Positive blood culture bottles (BacT/ALERT®, bioMérieux, Lyon, France) were mixed, and 4–5 mL of positive blood culture samples were withdrawn via a sterile syringe into yellow separating gel procoagulant tubes (BD Vacutainer®, Franklin Lakes, NJ, USA). The tubes were then centrifuged at 900×g for 10 min, and the supernatant was discarded. Grey-white colonies collected at the edge of the isolate were placed in 2 mL EP tubes. To these mixtures, 500 µL of deionized water and 1000 µL of 75% alcohol were added. The tubes were centrifuged at 6200–16,200×g for 2 min, and the supernatant was discarded. The centrifuged material was mixed with 50 µL of 70% formic acid solution. Then, 50 µL of acetonitrile was added, the mixture was mixed, and the mixture was centrifuged again at 6200–16,200×g for 2 min. The supernatant was collected for MALDI-TOF MS identification.

Identification and validation of pathogens in positive blood cultures

Pathogens with the same susceptibility pattern in patients during the same admission event (from admission to discharge) were isolated via the serum isolation gel method as described above and identified via MALDI-TOF MS, and E. coli and K. pneumoniae mass spectrometry peak mapping data were collected. The identified blood culture data were subjected to analysis in established DT, RF, GBM, XGBoost and ERT models with the objective of rapidly determining the presence of resistance to carbapenems (Fig. 2).

Fig. 2
figure 2

Pathogens in positive blood cultures were isolated by serum separation gels and identified by MALDI-TOF MS. The results of E. coli and K. pneumoniae identifications were analysed by machine learning models to rapidly predict susceptibility to carbapenems, thereby reducing laboratory turnaround time (this figure was produced in the biorender)

Full size image

Results

Machine learning models building data composition

In this study, a total of 640 E. coli cases and 444 K. pneumoniae cases of MALDI-TOF MS peak data were collected during the modelling process. Among the E. coli cases, 529 were carbapenem-sensitive Escherichia coli (82.66% of 640) and 111 were CREC (17.34% of 640). Among the K. pneumoniae cases, 301 were carbapenem-sensitive Klebsiella pneumoniae (67.79% of 444) and 143 were CRKP (32.21% of 444). The data identification rates for both E. coli and K. pneumoniae were > 90%. Tree-based models DT, RF, GBM, XGBoost, and ERT were constructed on the basis of the drug sensitivity results as binary labelling variables. A total of 149 E. coli and 127 K. pneumoniae were collected and identified from positive blood culture bottles. Among the E. coli, 132 were carbapenem-sensitive E. coli (88.59% of 149) and 17 were CREC (11.41% of 149). Among the K. pneumoniae, 101 were carbapenem-sensitive K. pneumoniae (79.53% of 127) and 26 were CRKP (20.47% of 127).

Escherichia coli machine learning models

Five tree-based machine learning models were trained on the data, and all models had AUROC curves exceeding 0.95 and ten-fold cross validation scores exceeding 0.95. All models were able to achieve a prediction accuracy of more than 0.86 for blood culture test data (Table 1). RF was the best overall performer among them, and interpreting the RF model data with a SHAP plot (Fig. 3), the importance of the top ten feature peaks were, in order, 7,870-7,879 m/z, 4,360-4,369 m/z, 6,250-6,259 m/z, 9,220-9,229 m/z, 8,870-8,879 m/z, 5,340-5,349 m/z, 7,150-7,159 m/z, 5,380-5,389 m/z, 6,270-6,279 m/z, and 3,930-3,939 m/z.

Table 1 Evaluation metrics for tree-based machine learning models of E. coli and predictive accuracy of blood cultures
Full size table
Fig. 3
figure 3

This is a Beeswarm plot of the Escherichia coli RF model, where the colour of each test spectrum indicates a data point, with red indicating high eigenvalues and blue indicating low eigenvalues. The vertical axis (Y-axis) represents the top ten features that have the greatest impact on RF prediction, and the horizontal axis (X-axis) represents the distribution of Shapley values and their impact on model output. Among them, 7,870-7,879 m/z is the feature peak that has the greatest impact on carbapenem-resistant Escherichia coli detection

Full size image

Klebsiella pneumoniae machine learning models

The five tree-based machine learning models exhibited slightly poor training efficacy for decision tree, which was considered weak evaluators. The other four models achieved an AUROC of 0.90 or greater, with ten-fold cross validation scores of 0.80 or greater, and predictive accuracies between 0.76 and 0.86 for blood cultures (Table 2). RF was the best overall performer among them, and interpreting the RF model data with a SHAP plot (Fig. 4), the importance of the top ten feature peaks were, in order, 4,920-4,929 m/z, 4,600-4,609 m/z, 2,260-2,269 m/z, 5,880-5,889 m/z, 4,450-4,459 m/z, 4,540-4,549 m/z, 5,930-5,939 m/z, 4,480-4,489 m/z, 10,580 − 10,589 m/z, and 9,840-9,849 m/z.

Table 2 Evaluation metrics for tree-based machine learning models of K. pneumoniae and predictive accuracy of blood cultures
Full size table
Fig. 4
figure 4

This is a Beeswarm plot of the Klebsiella pneumoniae RF model, where the colour of each test spectrum indicates a data point, with red indicating high eigenvalues and blue indicating low eigenvalues. The vertical axis (Y-axis) represents the top ten features that have the greatest impact on RF prediction, and the horizontal axis (X-axis) represents the distribution of Shapley values and their impact on model output. Among them, 4,920-4,929 m/z is the feature peak that has the greatest impact on carbapenem-resistant Klebsiella pneumoniae detection

Full size image

Discussion

Rapid detection of pathogens in blood cultures is crucial for combating infections. Reducing the detection cycle for identified bacteria enables early clinical decision-making and intervention [15]. The current method of relying on culture for pathogen identification followed by in vitro AST is the most established, has been extensively evaluated, is inexpensive, and is easy to perform. The limitations of culture are obvious; the different times required for the growth of each pathogen and the different technical requirements of culture can result in significantly longer turnaround times in the laboratory. Current nucleic acid-based methods for the direct detection of positive blood cultures (multiplex PCR, fluorescence in situ hybridization, and polymerase chain reaction/electrospray mass spectrometry) can reduce the detection time to less than 8 h, but the assay requires special instruments and reagents [16]. However, the test requires special instruments and reagents. Additionally, immunochromatographic tests offer rapid detection of carbapenem resistance genes in positive blood cultures, albeit at a relatively high cost [17]. To meet the demand for antimicrobial susceptibility testing of bloodstream infection pathogens, various rapid techniques have been developed, including Alfred 60 ASTTM, VITEK® REVEALTM, dRASTTM, ASTar®, Fastinov®, QuickMIC®, ResistellTM, and LifeScale. These methods aim to rapidly predict the antimicrobial susceptibility testing of bacteria causing bloodstream infections [18].

The use of MALDI-TOF MS enables the identification of bacteria without strain passaging cultures, the comparison of protein profiles obtained from bacterial or fungal samples with databases by mass spectrometry, and direct use of AST in purified or enriched microbial samples obtained from positive blood cultures [19]. Machine learning techniques can reveal novel or unknown information embedded in MALDI-TOF mass spectra. This information has been shown to be useful for species identification and differentiation, especially those that are phylogenetically close and sublineages of species [20]. Furthermore, antimicrobial resistance can be detected by analysing peak patterns or mass peak distributions in bacterial MALDI-TOF mass spectra [21]. In this study, a classification model constructed on the basis of MALDI-TOF data of E. coli and K. pneumoniae was applied for the detection of overall positive blood cultures, which can reduce the time to approximately 0.5 h.

Although the algorithms and modelling approaches differed among the tree models, the RF model was the best among the models developed in this study in terms of all evaluation metrics and was more accurate in predicting the blood culture validation set, with no significant overfitting/underfitting of the model. The detection efficacy of the models could basically meet the detection requirements, but the model efficacy of K. pneumoniae in this study was significantly worse than that of E. coli under the same data processing method, and the authors believed that this was mainly because some of the strains of K. pneumoniae presented a high mucus phenotype [22]. The high mucus phenotype can easily affect the laser excitation effect of the laser by unevenly applying the target when coating plates. As a result, the MALDI-TOF data of K. pneumoniae had a low pass rate and many stray peaks, which affected the final modelling results.

Tree modelling, as the most widely used machine learning model, is based on TreeExplainer, which provides fast local interpretation and guarantees consistency. It links theory to practice by reducing the complexity of exact Shapley value computation from exponential time to polynomial time [23]. Thus, tree-based models have some interpretability, are not necessarily worse than deep learning for processing data. In this study, we identified ten important feature peaks for the RF models of E. coli and K. pneumoniae, each, which significantly enhanced the transparency of the respective machine learning model algorithms. Among them, 7,870-7,879 m/z is the feature peak with the greatest impact on carbapenem-resistant E. coli detection, while 4,920-4,929 m/z is the peak with the greatest impact on carbapenem-resistant K. pneumoniae detection.

Both CREC and CRKP were better detected in this study on the basis of the tree-based model. For the tree-based model against E. coli, the model was able to achieve an AUROC of 0.95-1.00, with an accuracy of 0.86–0.92 on independent positive blood culture test data. For K. pneumoniae, each tree-based model had an AUROC of 0.78–0.95, with 0.76–0.86 accuracy on independent positive blood culture data. Reports of analysing MALDI-TOF data with machine learning algorithms are not uncommon in the literature. Gato E et al. [24] achieved 97.83% prediction of the CRKP with the constructed RF model. Jiaxin Y et al. [25] used the GBM algorithm to directly model the blood culture data of K. pneumoniae, and the AUROC of the model was 0.828. Yuming Z et al. [26] used a neural network algorithm to detect the CRKP, with an AUROC of 0.90. Although all of the above models achieved good detection results, none of them were tested via external independent data. While a machine learning model performs well on modelling data, this does not guarantee that it will be equally effective on new data. Therefore, we use an independent test set to validate the generalization ability of the model and ensure that it maintains efficient detection performance for other datasets as well.

This study is just an exploration of the application of machine learning algorithms to MALDI-TOF MS for the detection of Enterobacteriaceae susceptibility to carbapenems, and we must acknowledge to the limitations of this study. Firstly there was no duplication of the biology and technicalities of mass spectrometry identification, only qualifying mass spectrometry data were collected for modelling purposes. The accuracy of the study depends on the quality of the underlying data, including MALDI-TOF MS and antimicrobial susceptibility testing. Machine learning models also have inherent uncertainties and different parameters have different effects on the model. Secondly the data collection was limited to a single-centre study in one hospital and therefore does not accurately reflect the diversity of resistance patterns in different hospital settings or in different geographical locations. Future studies will focus on expanding the dataset to include more diverse samples from different geographic and clinical settings. These rigorous steps will significantly advance the field of antibiotic resistance detection and treatment, ensuring a more reliable and comprehensive diagnostic tool.

Conclusion

This study confirms that tree-based machine learning can be used to detect the susceptibility of E. coli and K. pneumoniae to carbapenems in positive blood cultures. This approach provides a cost-effective and rapid method for identifying carbapenem resistance, allowing rapid clinical decisions to be made in a short period of time. While challenges remain, these findings represent an important step forward in the fight against antibiotic resistance and offer new ways to improve diagnostic accuracy and provide treatment strategies.

Data availability

The data and python code supporting the results of this study are available from the corresponding author, Jiahong Ma.

References

  1. Rudd KE, Johnson SC, Agesa KM, et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the global burden of Disease Study[J]. Lancet. 2020;395(10219):200–11. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/S0140-6736(19)32989-7.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Goto M, Al-Hasan MN. Overall burden of bloodstream infection and nosocomial bloodstream infection in North America and Europe[J]. Clin Microbiol Infect. 2013;19(6):501–9. https://doiorg.publicaciones.saludcastillayleon.es/10.1111/1469-0691.12195.

    Article  CAS  PubMed  Google Scholar 

  3. Kern WV, Rieg S. Burden of bacterial bloodstream infection-a brief update on epidemiology and significance of multidrug-resistant pathogens[J]. Clin Microbiol Infect. 2020;26(2):151–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.cmi.2019.10.031.

    Article  CAS  PubMed  Google Scholar 

  4. Lamy B, Dargère S, Arendrup MC, et al. How to optimize the use of blood cultures for the diagnosis of Bloodstream infections? A state-of-the Art[J]. Front Microbiol. 2016;7:697. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fmicb.2016.00697.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Lawandi A. Bloodstream infections in Sepsis: better the Devil you Know[J]. Crit Care Med. 2023;51(9):1261–3. https://doiorg.publicaciones.saludcastillayleon.es/10.1097/CCM.0000000000005918.

    Article  PubMed  Google Scholar 

  6. Diekema DJ, Hsueh PR, Mendes RE, et al. The Microbiology of Bloodstream infection: 20-Year trends from the SENTRY Antimicrobial Surveillance Program [J]. Antimicrob Agents Chemother. 2019;63(7):e00355–19. https://doiorg.publicaciones.saludcastillayleon.es/10.1128/AAC.00355-19.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. WHO Bacterial Priority Pathogens List. 2024: Bacterial pathogens of public health importance to guide research, development and strategies to prevent and control antimicrobial resistance. Geneva: World Health Organization; 2024. Licence: CC BY-NC-SA 3.0 IGO.

  8. Rhee C, Kadri SS, Dekker JP, et al. Prevalence of antibiotic-resistant pathogens in Culture-Proven Sepsis and outcomes Associated with inadequate and broad-spectrum empiric antibiotic Use[J]. JAMA Netw Open. 2020;3(4):e202899. https://doiorg.publicaciones.saludcastillayleon.es/10.1001/jamanetworkopen.2020.2899.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Ackerman MH, Ahrens T, Kelly J, et al. Sepsis[J]. Crit Care Nurs Clin North Am. 2021;33(4):407–18. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.cnc.2021.08.003.

    Article  PubMed  Google Scholar 

  10. Tsuchida S, Umemura H, Nakayama T. Current status of Matrix-assisted laser Desorption/Ionization-Time-of-flight Mass Spectrometry (MALDI-TOF MS) in Clinical Diagnostic Microbiology[J]. Molecules. 2020;25(20):4775. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/molecules25204775.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Elbehiry A, Aldubaib M, Abalkhail A, et al. How MALDI-TOF Mass Spectrometry Technology contributes to microbial infection control in Healthcare Settings[J]. Vaccines (Basel). 2022;10(11):1881. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/vaccines10111881.

    Article  CAS  PubMed  Google Scholar 

  12. Kassim A, Pflüger V, Premji Z, et al. Comparison of biomarker based Matrix assisted laser desorption ionization-time of Flight Mass Spectrometry (MALDI -TOF MS) and conventional methods in the identification of clinically relevant bacteria and yeast[J]. BMC Microbiol. 2017;17(1):128. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12866-017-1037-z.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Parimbelli E, Buonocore TM, Nicora G, et al. Why did AI get this one wrong? - tree-based explanations of machine learning model predictions[J]. Artif Intell Med. 2023;135:102471. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.artmed.2022.102471.

    Article  PubMed  Google Scholar 

  14. Rai S, Dash D, Agarwal N. Introducing the new face of CLSI M100 in 2023: an explanatory review[J]. Indian J Med Microbiol. 2023;46:100432.

    Article  PubMed  Google Scholar 

  15. Sherwin R, Winters ME, Vilke GM, et al. Does early and Appropriate Antibiotic Administration Improve Mortality in Emergency Department patients with severe Sepsis or septic shock? [J]. J Emerg Med. 2017;53(4):588–95. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.jemermed.2016.12.009.

    Article  PubMed  Google Scholar 

  16. Opota O, Jaton K, Greub G. Microbial diagnosis of bloodstream infection: towards molecular diagnosis directly from blood[J]. Clin Microbiol Infect. 2015;21(4):323–31. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.cmi.2015.02.005.

    Article  CAS  PubMed  Google Scholar 

  17. de Lima AV, de Oliveira LK. Performance of two immunochromatographic tests in detecting KPC variants conferring resistance to ceftazidime -avibactam[J]. Eur J Clin Microbiol Infect Dis. 2024;43(6):1251–3. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s10096-024-04815-w.

    Article  CAS  PubMed  Google Scholar 

  18. Liborio MP, Harris P, Ravi C, et al. Getting up to speed: Rapid Pathogen and Antimicrobial Resistance Diagnostics in Sepsis[J]. Microorganisms. 2024;12(9):1824. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/microorganisms12091824.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Costa A, Figueroa-Espinosa R, Martínez JA, et al. MALDI-TOF MS-Based KPC Direct Detection from patients’ positive blood culture bottles, short-term cultures, and colonies at the Hospital[J]. Pathogens. 2023;12(7):865. https://doiorg.publicaciones.saludcastillayleon.es/10.3390/pathogens12070865.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Burckhardt I, Zimmermann S. Susceptibility testing of Bacteria using Maldi-Tof Mass Spectrometry[J]. Front Microbiol. 2018;9:1744. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fmicb.2018.01744.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Hrabák J, Chudácková E, Walková R. Matrix-assisted laser desorption ionization-time of flight (maldi-tof) mass spectrometry for detection of antibiotic resistance mechanisms: from research to routine diagnosis[J]. Clin Microbiol Rev. 2013;26(1):103–14. https://doiorg.publicaciones.saludcastillayleon.es/10.1128/CMR.00058-12.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Fang CT, Chuang YP, Shun CT, et al. A novel virulence gene in Klebsiella pneumoniae strains causing primary liver abscess and septic metastatic complications[J]. J Exp Med. 2004;199(5):697–705. https://doiorg.publicaciones.saludcastillayleon.es/10.1084/jem.20030857.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Lundberg SM, Erion G, Chen H, et al. From local explanations to Global understanding with explainable AI for Trees[J]. Nat Mach Intell. 2020;2(1):56–67. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/s42256-019-0138-9.

    Article  PubMed  PubMed Central  Google Scholar 

  24. Gato E, Arroyo MJ, Méndez G, et al. Direct detection of carbapenemase-producing Klebsiella pneumoniae by MALDI-TOF analysis of full Spectra applying machine Learning[J]. J Clin Microbiol. 2023;61(6):e0175122. https://doiorg.publicaciones.saludcastillayleon.es/10.1128/jcm.01751-22.

    Article  CAS  PubMed  Google Scholar 

  25. Yu J, Lin HH, Tseng KH, et al. Prediction of methicillin-resistant Staphylococcus aureus and carbapenem-resistant Klebsiella pneumoniae from flagged blood cultures by combining rapid Sepsityper MALDI-TOF mass spectrometry with machine learning[J]. Int J Antimicrob Agents. 2023;62(6):106994. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.ijantimicag.2023.106994.

    Article  CAS  PubMed  Google Scholar 

  26. Zhang YM, Tsao MF, Chang CY, et al. Rapid identification of carbapenem-resistant Klebsiella pneumoniae based on matrix-assisted laser desorption ionization time-of-flight mass spectrometry and an artificial neural network model[J]. J Biomed Sci. 2023;30(1):25. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12929-023-00918-2.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

We thank Prof. Xu for her painstaking contribution to this study.

Funding

No funding was received for this study.

Author information

Authors and Affiliations

  1. Department of Clinical Laboratory, Zhejiang Rong Jun Hospital, Jiaxing, 314000, China

    Xiaobo Xu, Zhaofeng Wang, Erjie Lu, Tao Lin, Hengchao Du, Zhongfei Li & Jiahong Ma

Authors
  1. Xiaobo Xu
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Zhaofeng Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Erjie Lu
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Tao Lin
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Hengchao Du
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Zhongfei Li
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Jiahong Ma
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

Xu was responsible for the study’s conception, design, experimentation, figure creation, data analysis, and manuscript preparation.Wang and Lu contributed to data collection, while Lin, Du, and Li participated in data analysis. Ma offered overall support for the project.

Corresponding author

Correspondence to Jiahong Ma.

Ethics declarations

Ethical approval

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Xu, X., Wang, Z., Lu, E. et al. Rapid detection of carbapenem-resistant Escherichia coli and carbapenem-resistant Klebsiella pneumoniae in positive blood cultures via MALDI-TOF MS and tree-based machine learning models. BMC Microbiol 25, 44 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12866-025-03755-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12866-025-03755-5

Keywords

BMC Microbiology

ISSN: 1471-2180

Contact us