Bacterial species identification using MALDI-TOF mass spectrometry and machine learning techniques: A large-scale...

Area of Science:

• Microbiology
• Computational Biology
• Analytical Chemistry

Background:

• Machine learning is widely used for bacterial identification via MALDI-TOF mass spectrometry.
• Existing studies often use limited datasets and traditional methods, lacking comprehensive benchmarking.
• Large-scale validation is needed to assess method performance across diverse bacterial species and identification scenarios.

Purpose of the Study:

• To benchmark a wide range of machine learning methods for bacterial identification using MALDI-TOF mass spectrometry on an unprecedented scale.
• To compare identification performance for novel biological replicates, novel strains, and novel species.
• To evaluate the preservation of taxonomic information in MALDI-TOF mass spectrometry data and explore novel methods for new species detection.

Main Methods:

• Benchmarking machine learning algorithms on a dataset of nearly 100,000 MALDI-TOF mass spectra from over 1000 bacterial species.
• Comparing identification accuracy across three scenarios: novel replicates, novel strains, and novel species.
• Utilizing hierarchical classification to assess taxonomic information representation.
• Applying neural networks with Monte Carlo dropout for novel species identification.

Main Results:

• Acceptable bacterial identification rates were achieved across all three scenarios (novel replicates, strains, species), though generally lower than reported in smaller-scale studies.
• Hierarchical classification indicated that taxonomic information is not consistently well-preserved in MALDI-TOF mass spectrometry data.
• Neural networks with Monte Carlo dropout demonstrated potential for novel species detection in the challenging novel species identification scenario.

Conclusions:

• Large-scale benchmarking reveals the practical performance limitations of machine learning methods for MALDI-TOF based bacterial identification.
• MALDI-TOF mass spectrometry data may not reliably preserve detailed taxonomic information for hierarchical classification.
• Advanced methods like neural networks with Monte Carlo dropout offer promising avenues for identifying previously uncharacterized bacterial species.