Rapid identification of salmonella serovars by using Raman spectroscopy and machine learning algorithm.

Area of Science:

• Food safety and microbiology
• Spectroscopy and analytical chemistry
• Machine learning in bioinformatics

Background:

• Foodborne illnesses, particularly Salmonella infections, pose a significant global public health challenge.
• Accurate and rapid identification of pathogenic bacteria is crucial for preventing outbreaks.
• Traditional methods for bacterial identification can be time-consuming and labor-intensive.

Purpose of the Study:

• To develop and evaluate a rapid method for identifying pathogenic Salmonella serotypes using Raman spectroscopy and machine learning.
• To compare the effectiveness of different laser wavelengths and spectral preprocessing techniques for Salmonella detection.
• To assess the performance of a convolutional neural network (CNN) model for classifying Salmonella serotypes.

Main Methods:

• Acquisition of Raman spectral data from three pathogenic Salmonella serotypes.
• Optimization of instrument parameters, including comparison of laser wavelengths (532, 638, and 785 nm).
• Evaluation of five spectral preprocessing methods (Savitzky-Golay smoothing, Multivariate Scatter Correction, Standard Normal Variate, Hilbert Transform) for noise reduction and signal enhancement.
• Application of a convolutional neural network (CNN) model for multi-classification of Salmonella serotypes.
• Performance evaluation using accuracy, precision, recall, and F1-score.

Main Results:

• The 532 nm laser wavelength was identified as the most effective for Salmonella detection.
• The combination of Savitzky-Golay smoothing (SG) and Standard Normal Variate (SNV) preprocessing yielded the highest accuracy.
• The CNN model achieved 98.7% accuracy on the training set and over 98.5% on the test set using the optimized SG+SNV preprocessing method.
• The developed method enabled accurate identification of Salmonella serotypes at the single-cell level.

Conclusions:

• Raman spectroscopy coupled with a CNN model provides a rapid and accurate method for identifying Salmonella serotypes.
• The SG+SNV spectral preprocessing significantly enhances the predictive power of the CNN model.
• This approach holds substantial potential for distinguishing between different pathogenic bacterial serotypes and closely related species, aiding in the prevention of foodborne illness outbreaks.