Topological and kernel-based microbial phenotype prediction from MALDI-TOF mass spectra.

Area of Science:

• Microbiology
• Computational Biology
• Mass Spectrometry

Background:

• Matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry (MS) is standard for microbial identification.
• MALDI-TOF MS spectra hold potential for predicting phenotypes like antibiotic resistance.
• Current machine learning for MALDI-TOF MS phenotype prediction is nascent, with complex pre-processing and lacking uncertainty quantification.

Purpose of the Study:

• To develop a novel prediction method for antimicrobial resistance using MALDI-TOF mass spectra.
• To compare conventional spectral pre-processing with a new topological information approach.
• To introduce a robust similarity measure and classifier with uncertainty estimation for MALDI-TOF MS data.

Main Methods:

• A new pre-processing method exploiting topological information with a single parameter (number of peaks).
• Introduction of the peak information kernel (PIKE), a similarity measure for MALDI-TOF MS spectra.
• Combination of PIKE with a Gaussian process classifier for well-calibrated uncertainty estimates.

Main Results:

• The novel method accurately predicts antibiotic resistance for three key bacterial species.
• The approach outperforms existing methods and enhances security by rejecting out-of-distribution samples.
• The PIKE method provides reliable uncertainty quantification for predictions.

Conclusions:

• The developed method offers a more efficient and secure approach to predicting antimicrobial resistance from MALDI-TOF MS data.
• This tool has the potential to facilitate earlier and more precise antimicrobial treatment in clinical settings.
• The publicly available Python package (maldi_PIKE) promotes accessibility and further research.