Scholarlink

<p>The development of microfluidic-based Raman devices has enabled the acquisition of thousands of single-cell Raman spectra from complex microbial populations, offering new avenues for characterizing environmental bacteria. We proposed and validated the convolutional deep embedded clustering (CDEC) model, a clustering framework designed for the unsupervised classification of marine microbial Raman spectra. A four-stage analytical framework was employed to systematically assess the performance of the CDEC model, with increasing dataset complexity ranging from mock communities of pure bacterial cultures to natural microbial populations isolated from seawater. The CDEC algorithm consistently distinguished bacterial species across all stages, achieving an average accuracy of 97.27%, and surpassed the baseline methods in performance. In natural microbial communities, the model resolved eight distinct clusters with unique Raman spectral markers, revealing underlying metabolic heterogeneity. These results highlight the scalability and utility of the CDEC model for studying marine microbial diversity and characterizing microbial composition and metabolic traits at single-cell resolution.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Unsupervised classification of environmental marine microbes using a Raman spectra-based deep learning framework

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Fengyun Lü,
  • Yanmei Zhang,
  • Hongtao Liang,
  • Luyang Sun

摘要

The development of microfluidic-based Raman devices has enabled the acquisition of thousands of single-cell Raman spectra from complex microbial populations, offering new avenues for characterizing environmental bacteria. We proposed and validated the convolutional deep embedded clustering (CDEC) model, a clustering framework designed for the unsupervised classification of marine microbial Raman spectra. A four-stage analytical framework was employed to systematically assess the performance of the CDEC model, with increasing dataset complexity ranging from mock communities of pure bacterial cultures to natural microbial populations isolated from seawater. The CDEC algorithm consistently distinguished bacterial species across all stages, achieving an average accuracy of 97.27%, and surpassed the baseline methods in performance. In natural microbial communities, the model resolved eight distinct clusters with unique Raman spectral markers, revealing underlying metabolic heterogeneity. These results highlight the scalability and utility of the CDEC model for studying marine microbial diversity and characterizing microbial composition and metabolic traits at single-cell resolution.