Domain-Level Classification of Archaea and Bacteria Using AI-Assisted Single-Cell Raman Spectroscopy.

Area of Science:

• Microbiology
• Spectroscopy
• Machine Learning

Background:

• Archaea and Bacteria are distinct prokaryotic domains with differing molecular and cellular structures.
• Archaea are found in extreme and moderate environments but are less characterized than bacteria due to culturing challenges.
• Metagenomic studies reveal archaea's widespread presence across various habitats, including the human microbiome.

Purpose of the Study:

• To develop a culture-independent method for discriminating between Archaea and Bacteria at the single-cell level.
• To create a reliable classifier for identifying archaeal species using Raman spectroscopy and machine learning.
• To provide a tool for studying unculturable or low-abundance archaeal populations.

Main Methods:

• Utilized Raman spectroscopy to collect spectral data from 22 prokaryotic species (11 archaea, 11 bacteria).
• Developed an Archaea-Bacteria (AB) classifier using the LightGBM machine learning algorithm.
• Compared the performance of the LightGBM classifier against convolutional neural networks with transfer learning.

Main Results:

• The LightGBM-based AB classification model achieved an average accuracy of 89.1% and a sensitivity of 98.1%.
• The model demonstrated high performance with minimal data size and preprocessing.
• The developed method offers a robust framework for archaeal detection.

Conclusions:

• The Raman spectroscopy and machine learning approach provides an effective means for single-cell discrimination of Archaea and Bacteria.
• This method enhances the microbiological toolkit for identifying and studying archaea, especially in complex communities.
• The technique is valuable for investigating archaeal populations that are difficult to culture or are present in low abundance.