Negative-Ion Mode MALDI-TOF MS Combined with Machine Learning for the Rapid Identification of Colistin-Resistant <i>E. cloacae</i> Complex

High Resolution Image Download MS PowerPoint Slide The Enterobacter cloacae complex (ECC) comprises major nosocomial pathogens and is increasingly resistant to last-line agents such as colistin. Here, we developed a rapid workflow that integrates negative-ion mode MALDI-TOF mass spectrometry with machine learning to identify colistin-resistant ECC (COL-R-ECC). We analyzed 267 clinical ECC isolates; species identification was performed by MALDI-TOF MS and verified by whole-genome sequencing. A one-dimensional convolutional neural network (1D-CNN) with a squeeze-and-excitation (SE) module was trained on spectra from 217 isolates and evaluated on an independent external cohort of 50 isolates. To enhance analytical specificity, we applied a standardized lipid extraction protocol and focused on lipid A–enriched signals ( m / z 1500–3000). Spectra were binned at 0.1 Da and smoothed using a Savitzky–Golay filter. The optimized model achieved 95.5% accuracy (AUROC = 0.986; F1 = 0.943) in internal testing and 96.0% accuracy (F1 = 0.960) in external validation, outperforming conventional machine-learning baselines. SHAP analysis highlighted 30 lipid-associated features that contributed most to the predictions, providing interpretable clues to potential resistance mechanisms. Overall, this workflow enables COL-R-ECC identification within ∼1 h from colony processing, substantially faster than standard susceptibility testing, and may support earlier targeted therapy and infection control.