Direct antimicrobial resistance prediction from clinical MALDI-TOF mass spectra using machine learning

Area of Science:

• Microbiology
• Infectious Diseases
• Computational Biology

Background:

• Early antimicrobial treatment is crucial for infection outcomes and preventing resistance.
• Current antimicrobial resistance testing methods (culture-based) are time-consuming, often taking up to 72 hours.
• Rapid determination of antimicrobial susceptibility is needed for timely and effective treatment selection.

Purpose of the Study:

• To develop and validate a novel machine learning (ML) approach for predicting antimicrobial resistance directly from MALDI-TOF mass spectra.
• To accelerate the process of antimicrobial resistance determination.
• To assess the potential clinical impact of ML-based antimicrobial resistance prediction.

Main Methods:

• Development of ML classifiers trained on a large, publicly available database of MALDI-TOF mass spectra linked to antimicrobial susceptibility phenotypes.
• The dataset comprises over 300,000 mass spectra and 750,000 resistance phenotypes from four institutions.
• Validation was performed on clinically relevant pathogens like Staphylococcus aureus, Escherichia coli, and Klebsiella pneumoniae.

Main Results:

• The ML approach demonstrated potential in predicting antimicrobial resistance from MALDI-TOF mass spectra.
• Achieved areas under the receiver operating characteristic curve (AUC) of 0.80 for S. aureus, 0.74 for E. coli, and 0.74 for K. pneumoniae.
• A retrospective clinical case study indicated that this approach could have altered treatment in 9 out of 63 patients, with 89% of these changes being beneficial.

Conclusions:

• Machine learning applied to MALDI-TOF mass spectra offers a promising method for rapid antimicrobial resistance determination.
• This technology has the potential to significantly accelerate clinical decision-making for antibiotic treatment.
• The approach supports improved treatment optimization and antibiotic stewardship efforts.