Rapid Identification of Methicillin-Resistant Staphylococcus aureus Using MALDI-TOF MS and Machine Learning from over 20,000 Clinical Isolates.

Rapidly identifying methicillin-resistant Staphylococcus aureus (MRSA) with high integration in the current workflow is critical in clinical practices. We proposed a matrix-assisted laser desorption/ionization-time-of-flight mass spectrometry (MALDI-TOF MS)-based machine learning model for rapid MRSA prediction. The model was evaluated on a prospective test and four external clinical sites. For the data set comprising 20,359 clinical isolates, the area under the receiver operating curve of the classification model was 0.78 to 0.88. These results were further interpreted using shapely additive explanations and presented using the pseudogel method. The important MRSA feature, 6,590 to 6,599, was identified as a UPF0337 protein SACOL1680 with a lower binding affinity or no docking results compared with UPF0337 protein SA1452, which is mainly detected in methicillin-susceptible S. aureus. Our MALDI-TOF MS-based machine learning model for rapid MRSA identification can be easily integrated into the current clinical workflows and can further support physicians in prescribing proper antibiotic treatments. Over 20,000 clinical MSSA and MRSA isolates were collected to build a machine learning (ML) model to identify MSSA/MRSA and their markers. This model was tested across four external clinical sites to ensure the model's usability. We report the first discovery and validation of MRSA markers on the largest scale of clinical MSSA and MRSA isolates collected to date, covering five different clinical sites. Our developed approach for the rapid identification of MSSA and MRSA can be highly integrated into the current workflows.