Rapid Detection of NDM-Producing Carbapenem-Resistant Escherichia coli Using MALDI-TOF MS Combined with Machine Learning Techniques

Background Carbapenem-resistant Escherichia coli (CREC) producing New Delhi metallo-beta-lactamase (NDM) poses a substantial threat to global public health. Prompt detection of CREC is essential for effective patient management and to curb the spread of resistance. Objective This study aims to establish a rapid and reliable detection method for NDM-producing CREC using matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) in conjunction with machine learning (ML) techniques. Methods Clinical isolates of E. coli were collected from the First Hospital of Jilin University’s clinical laboratory from August 2018 to December 2022. PCR was employed to detect 23 prevalent resistance genes in the isolates. Protein profiles were analyzed by MALDI-TOF MS with a cooperative matrix consisting of (E)-propyl α-cyano-4-hydroxylcinnamylate (CHCA-C3) and α-cyano-4-hydroxycinnamic acid (CHCA). By comparative analysis of nine ML algorithms, the optimal Random Forest (RF) algorithm was identified and used to find the NDM specific ion peaks. Furthermore, the NDM gene was cloned into the standard E. coli strain ATCC 25922 to verify the reliability of these specific peaks. Results A total of 154 E. coli strains were categorized into three groups: 45 CREC strains, 58 extended-spectrum beta-lactamase (ESBL)-producing strains, and 51 non-CREC, non-ESBL strains. PCR amplification identified 12 resistance genes. The established ML model with RF algorithm exhibits excellent discriminative ability, achieving an AUC of 0.993 and an AP of 0.997. Four specific peaks ( m/z 5132, m/z 5209, m/z 6350, and m/z 6371) of NDM were identified using the cooperative matrix. In contrast, the traditional CHCA matrix revealed no specific peaks. Furthermore, the specific peaks at m/z 5132 and m/z 5209 were detected in the MALDI-TOF MS spectra of the recombinant E. coli strain ATCC 25922-PZY01/NDM, which directly correlated with the expression of the NDM gene. Conclusion The integration of MALDI-TOF MS with ML techniques, facilitated by the cooperative matrix approach, has enabled the swift and precise identification of NDM-producing CREC. This advancement significantly enhances the effectiveness of clinical management and the control of hospital-acquired infections.