Archaea and Bacteria are two fundamentally distinct domains of life that share prokaryotic traits, yet differ markedly in molecular and cellular architecture. While many archaeal species identified thus far have been found in extreme environments, recent metagenomic studies have revealed their widespread presence in moderate habitats, including soils, oceans, and even the human microbiome. However, archaea remain less well characterized than bacteria, largely due to the technical challenges associated with culturing and identifying these microorganisms. In this study, we present a culture-independent method for discriminating archaea from bacteria at the single-cell level using Raman spectroscopy combined with machine learning. We constructed a Raman spectral data set comprising 22 prokaryotic species (11 archaea and 11 bacteria) and developed a domain-level Archaea–Bacteria (AB) classifier using the LightGBM tree-based machine learning algorithm. Our AB classification model achieved an average classification accuracy of 89.1% and a sensitivity of 98.1% on eight representative species (including two independent held-out test species) with minimal data size and preprocessing. We also compared its performance to convolutional neural networks with transfer learning, a widely used deep learning approach. Our method provides a robust analytical framework for archaeal detection and represents a valuable addition to the microbiological toolkit, particularly for studying unculturable or low-abundance archaeal populations in complex microbial communities.
Kanno et al. (Tue,) studied this question.