Deep Learning for Rapid Identification of Microbes Using Metabolomics Profiles

Area of Science:

• Microbiology
• Computational Biology
• Analytical Chemistry

Background:

• Rapid detection of viable microbes is crucial for microbial food safety.
• Current methods face challenges in speed and accuracy.
• Metabolomics offers a rich data source for microbial analysis.

Purpose of the Study:

• To apply deep learning algorithms for rapid microbial detection.
• To utilize metabolomics data for identifying pathogenic and non-pathogenic microbes.
• To compare the performance of artificial neural networks (ANN) against other machine learning models.

Main Methods:

• Microbes were incubated for 4 hours in a defined medium.
• 1D proton nuclear magnetic resonance (NMR) spectroscopy was used for data acquisition.
• Untargeted metabolomics analysis involved spectral binning and ANN model training.

Main Results:

• ANN models achieved an average accuracy of 99.2% in predicting unknown microbes.
• Feature selection enhanced ANN performance.
• ANN consistently outperformed random forest and support vector machine models.

Conclusions:

• Deep learning, specifically ANNs, shows significant potential for analyzing metabolomics data.
• This approach enables rapid and accurate microbial detection.
• The findings highlight the utility of ANNs in microbial food safety and broader metabolomics applications.