Rapid prediction of vancomycin-resistant Enterococcus faecium using MALDI-TOF mass spectrometry and machine learning

Background Vancomycin-resistant Enterococcus faecium (VRE fm ) is a World Health Organization priority pathogen, yet conventional phenotypic susceptibility testing requires up to 72 h, delaying targeted antimicrobial therapy. This study aimed to develop an interpretable and rapid machine learning classifier to predict VRE fm using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) spectra. Methods We retrospectively analyzed 268 clinical E. faecium isolates (100 VRE fm and 168 vancomycin-susceptible E. faecium ) from Tangshan Gongren Hospital, a tertiary hospital in northern China (2020–2025). Spectra were preprocessed with smoothing, baseline removal, and 5-Da binning (range 2,000–20,000 Da). Extreme gradient boosting recursive feature elimination selected 40 discriminative mass-to-charge ratio features. To enhance model generalizability across diverse bacterial lineages, the training set combined local isolates from 2020 to 2024 with the multi-center DRIAMS repository (subsets A–D). Five classifiers were trained under stratified 5-fold cross-validation and temporally validated on the independent 2025 local isolates. Model discrimination, calibration, and clinical utility were evaluated using the area under the receiver operating characteristic curve (AUROC), Brier score, and decision curve analysis. Results The k-nearest neighbors classifier achieved optimal temporal validation performance (area under the receiver operating characteristic curve 0.90, F1 score 0.683, Brier score 0.185). Hybrid training configurations combining local data with pooled DRIAMS subsets retained clinically useful discrimination (AUROC > 0.80), whereas external-only models performed at chance level (AUROC ≈ 0.50). Conclusion MALDI-TOF MS combined with machine learning enables rapid, interpretable prediction of vancomycin resistance in E. faecium . A hybrid local–multi-center training strategy offers a pragmatic solution for laboratories with limited local sample availability, facilitating clinical deployment of spectral-based antimicrobial resistance surveillance.