1：Experimental Design and Method Selection

The study employed a combination of laser tweezers Raman spectroscopy (LTRS) and a convolutional neural network (ConvNet) for microbial identification. The LTRS system was used to acquire Raman spectra of single microbial cells, and the ConvNet was trained to classify these spectra.

2：Sample Selection and Data Sources

Fourteen microbial species, including bacteria, archaea, and fungi, were selected. For each species, more than 300 single-cell Raman spectra were measured under various culture conditions to ensure diversity.

3：List of Experimental Equipment and Materials

Laser tweezers Raman spectroscopy system with a 785-nm laser, confocal fluorescence microscopy, optical tweezers, Raman spectrometer (Acton SP2300; Princeton Instruments), and thermoelectric-cooled charge-coupled device (PIXIS: 400BD).

4：Experimental Procedures and Operational Workflow

Single microbial cells were trapped and their Raman spectra were measured using the LTRS system. The spectra were pre-processed to remove cosmic rays and fluorescence background, then normalized. The ConvNet model was trained using these spectra and validated through 10-fold cross-validation.

5：Data Analysis Methods

The ConvNet model was used for classification, with performance evaluated using accuracy, error rate, and validation loss. An occlusion-based Raman spectra feature extraction (ORSFE) method was developed to identify significant Raman features for classification.