Efficiently Predicting Vancomycin Resistance of Enterococcus Faecium From MALDI-TOF MS Spectra Using a Deep...

Area of Science:

• Microbiology
• Analytical Chemistry
• Bioinformatics

Background:

• Matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry (MS) is crucial for microbial identification.
• Current methods using single peaks for antibiotic resistance prediction in bacteria have limited accuracy.
• Machine learning offers potential for analyzing complex MALDI-TOF MS spectral patterns.

Purpose of the Study:

• To develop a deep learning model for rapid and accurate detection of vancomycin-resistant Enterococcus faecium (VREfm).
• To leverage the complete information from MALDI-TOF MS spectra for enhanced bacterial identification and resistance prediction.
• To identify key spectral features contributing to VREfm detection.

Main Methods:

• Development of a convolutional neural network (CNN) model to analyze whole MALDI-TOF MS spectra.
• Application of the CNN model to clinical samples for VREfm detection.
• Utilizing score-class activation mapping (CAM) to interpret CNN model predictions and identify important spectral features.

Main Results:

• The CNN model achieved good classification performance, with an average area under the receiver operating characteristic curve (AUROC) of 0.887 on external validation data.
• The CAM method successfully identified discriminative signals within the MS spectra crucial for VREfm detection.
• The study demonstrated the effectiveness of using complete spectral information for improved prediction.

Conclusions:

• A CNN-based approach effectively utilizes comprehensive MALDI-TOF MS data for rapid VREfm detection.
• This deep learning method provides a practical and accurate tool for identifying antibiotic-resistant bacterial strains.
• The findings highlight the potential of AI in enhancing microbial diagnostics and combating antimicrobial resistance.