AI-assisted MALDI-TOF MS for identifying carbapenem resistance in clinical Acinetobacter baumannii isolates

Abstract

ABSTRACT Carbapenem-resistant Acinetobacter baumannii (CRAB) is one of the most critical public health threats worldwide due to its high infection rates, substantial mortality, and limited therapeutic choices. As CRAB infections are frequently multidrug-resistant, rapid and accurate determination of carbapenem susceptibility is essential for appropriate therapeutic decision-making. We established an integrated framework combining matrix-assisted laser desorption ionization–time-of-flight mass spectrometry (MALDI-TOF MS) with artificial intelligence (AI) to enable rapid prediction of carbapenem resistance in A. baumannii. A total of 191 clinical and surveillance isolates, including CRAB and carbapenem-susceptible A. baumannii (CSAB), were recovered from hospitalized patients and phenotypically characterized by standard minimum inhibitory concentration testing. MALDI-TOF MS spectra were subsequently acquired, and six AI models were developed and systematically investigated for predictive performance, among which the eXtreme Gradient Boosting (XGBoost) model achieved the highest discriminatory performance, distinguishing CRAB from CSAB with an accuracy of 96.36% and robust overall performance. Feature importance analysis of the XGBoost model revealed that its high predictive performance was driven partially by spectral features associated with horizontal gene transfer-related elements and membrane and transport-associated proteins, providing candidate molecular correlates of carbapenem resistance. Overall, these results show that AI-assisted MALDI-TOF MS enables rapid and accurate prediction of carbapenem resistance in A. baumannii and provides insights into the molecular features associated with resistance acquisition.