Development and validation of a predictive model for profiling antibiotic resistance phenotypes of Acinetobacter...

Acinetobacter baumannii is a significant pathogen responsible for healthcare-acquired infections (HAIs), posing challenges due to its rising resistance to multiple antibiotics. Traditional diagnostic methods like bacterial cultures and antibiotic susceptibility testing (AST) are slow, necessitating the development of faster alternatives. This study aimed to create a predictive model for antibiotic resistance phenotypes of A. baumannii using clinical MALDI-TOF mass spectrometry (MS) spectra combined with machine learning techniques. A total of 3644 A. baumannii strains were analysed for resistance patterns to nine antibiotic classes. Eight machine learning models were trained and evaluated, with the best model showing over 83% accuracy in predicting resistance to carbapenems, penicillins, and quinolones. The model performed especially well for imipenem and ceftazidime, with accuracies of 84.90% and 84.64%, respectively. Multi-center validation confirmed the model's robustness, achieving 81.68% and 80.72% accuracy for imipenem and ceftazidime. These findings demonstrate the potential of integrating machine learning with MALDI-TOF MS for rapid, accurate profiling of A. baumannii's antimicrobial resistance in clinical settings.