Deep Learning-Enhanced Raman Microspectroscopy Enables Rapid Microbial Classification and Captures Phylogenetic...

Area of Science:

• Microbiology
• Spectroscopy
• Bioinformatics

Background:

• Microbial classification and taxonomy are crucial for microbiology.
• Raman microspectroscopy provides rapid, non-destructive single-cell molecular fingerprints.
• Current Raman methods struggle with uncharacterized microorganisms.

Purpose of the Study:

• To develop a deep learning framework for microbial classification using single-cell Raman spectroscopy.
• To assess the accuracy and taxonomic congruence of the developed method.
• To provide a complementary tool for microbial taxonomy, especially for novel species.

Main Methods:

• Collected 6600 single-cell Raman spectra from 11 microbial species.
• Developed deep learning models, including a 1D-CNN, for feature extraction.
• Constructed a hierarchical clustering framework based on Raman features.
• Compared Raman-based classification trees with rRNA gene sequence-based phylogenetic trees.

Main Results:

• The 1D-CNN achieved 99.7% classification accuracy.
• The Raman hierarchical clustering tree showed strong concordance with phylogenetic structures.
• Independent validation with unknown strains confirmed accurate placement near phylogenetic relatives.

Conclusions:

• Single-cell Raman spectroscopy with deep learning is a viable alternative/complementary method for microbial taxonomy.
• This approach shows significant potential for classifying previously uncharacterized microorganisms.
• The Raman HC tree effectively reflects microbial evolutionary relationships.