Performance of a Machine Learning-Based Methicillin Resistance of Staphylococcus aureus Identification System Using MALDI-TOF MS and Comparison of the Accuracy according to SCCmec Types

Abstract

The prompt presumptive identification of methicillin-resistant Staphylococcus aureus (MRSA) using matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) can aid in early clinical management and infection control during routine bacterial identification procedures. This study applied a machine learning approach to MALDI-TOF peaks for the presumptive identification of MRSA and compared the accuracy according to staphylococcal cassette chromosome mec (SCCmec) types. We analyzed 194 S. aureus clinical isolates to evaluate the machine learning-based identification system (AMRQuest software, v.2.1, ASTA: Suwon, Korea), which was constructed with 359 S. aureus clinical isolates for the learning dataset. This system showed a sensitivity of 91.8%, specificity of 83.3%, and accuracy of 87.6% in distinguishing MRSA. For SCCmec II and IVA types, common MRSA types in a hospital context, the accuracy was 95.4% and 96.1%, respectively, while for the SCCmec IV type, it was 21.4%. The accuracy was 90.9% for methicillin-susceptible S. aureus. This presumptive MRSA identification system may be helpful for the management of patients before the performance of routine antimicrobial resistance testing. Further optimization of the machine learning model with more datasets could help achieve rapid identification of MRSA with less effort in routine clinical procedures using MALDI-TOF MS as an identification method.