Rapid Identification of Carbapenemase Subtypes in Klebsiella pneumoniae Using MALDI-TOF MS Combined With Convolutional Neural Networks

The global spread of carbapenem-resistant Klebsiella pneumoniae (CRKP) severely threatens public health, with its resistance profile highly dependent on carbapenemase subtypes, such as Klebsiella pneumoniae carbapenemase (KPC), New Delhi metallo-β-lactamase (NDM), and oxacillinase (OXA)-48. However, existing detection technologies have limitations such as insufficient sensitivity, high cost or complex operation. This study combines matrix-assisted laser desorption/ionization time-of-flight mass spectrometry with convolutional neural networks (CNNs) to establish a rapid and precise system for identifying carbapenemase subtypes in CRKP. An end-to-end CNN classification model was constructed by performing high-resolution mass spectrometry analysis on 205 clinical CRKP isolates (including 111 KPC–, 47 NDM– and 47 OXA-48–producing isolates), with 12 technical replicates per strain. The model employs a hierarchical convolutional architecture to automatically analyze full-spectrum domain features, achieving an overall accuracy of 96.1% in five-fold cross-validation, with a KPC subtype identification rate of 98.88%. Compared with traditional machine learning methods, such as random forest, support vector machine and AdaBoost, CNN demonstrated superior parameter efficiency and subtype discrimination capability under limited sample sizes. The technical process can be completed within 20 minutes after bacterial culture at a significantly lower cost than that of molecular detection. The modular design of the model supports the rapid integration of region-specific strain databases, providing a scalable solution to address the regional heterogeneity of carbapenemase distribution. This study confirms the translational potential of artificial intelligence–enhanced mass spectrometry technology in enzyme-targeting precision therapy and antibiotic resistance control, providing clinical microbiology laboratories with an alternative tool that offers timeliness, cost-effectiveness and diagnostic accuracy.