Neural network-enabled MALDI-TOF MS for rapid clinical identification of methicillin-resistant Staphylococcus aureus.

OBJECTIVE: To investigate the feasibility of utilizing neural network algorithm to analyze parameters from matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) for rapid identification of methicillin-resistant Staphylococcus aureus (MRSA). METHODS: Strain identification was performed using MALDI-TOF MS, and antimicrobial susceptibility testing was conducted with the VITEK 2 Compact AST-GP639 card. Mass spectrometry parameters were collected from 41 Staphylococcus aureus strains isolated from different patients and anatomical sites, including 20 MRSA and 21 methicillin-susceptible Staphylococcus aureus (MSSA) isolates. Twenty characteristic mass spectral features were screened using LASSO regression analysis. A neural network model was established using these 20 mass spectrometry parameters to rapidly discriminate MRSA. RESULTS: LASSO identified 20 discriminatory m/z features, with individual AUCs ranging from 0.825 to 0.982. The ANN model, evaluated by leave-one-out cross-validation, achieved an overall accuracy of 97.56%, with 95% sensitivity and 100% specificity. Compared to single characteristic mass spectral features, the discrimination efficiency of the ANN model was significantly improved, while demonstrating high consistency with classical methods. In addition, SHapley Additive explanations analysis revealed that characteristic features e.g. m/z 147.7004 and 39,767.2735 played major roles in model predictions. CONCLUSION: The neural network model based on MALDI-TOF MS parameters enables accurate discrimination between MRSA and MSSA. This approach offers rapid identification and cost-effectiveness, providing a novel strategy for clinical differentiation of MRSA and MSSA.