Performance of Classification Models of Toxins Based on Raman Spectroscopy Using Machine Learning Algorithms

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Biochemistry

Background:

• Rapid and accurate detection of protein toxins is critical for safeguarding public health.
• Existing detection methods may lack the speed or accuracy required for immediate threat assessment.
• Raman spectroscopy offers a sensitive and specific method for molecular analysis.

Purpose of the Study:

• To develop and validate a Raman spectroscopy-based method for the rapid detection and classification of protein toxins.
• To evaluate the effectiveness of various spectral preprocessing techniques and chemometric models for toxin identification.
• To establish robust classification models for distinguishing between different protein toxins and other proteins.

Main Methods:

• Raman spectra of protein toxins (abrin, ricin, SEB, BGT) and control proteins were acquired.
• Spectral data preprocessing involved multivariate scattering correction (MSC), Savitzky-Golay smoothing (SG), and wavelet transform (WT).
• Feature extraction was performed using principal component analysis (PCA), followed by classification using k-means, partial least squares discriminant analysis (PLS-DA), and partial least squares regression (PLSR).

Main Results:

• PCA score plots effectively clustered the four studied toxins and two other proteins.
• Spectra preprocessed with MSC and MSC-SG methods yielded the best classification performance with k-means.
• PLS-DA achieved 100% accuracy in classifying the two data types.
• PLSR demonstrated excellent classification and regression capabilities (accuracy = 100%, Rcv = 0.776), correctly identifying four toxins despite interference from two other proteins.

Conclusions:

• The developed Raman spectroscopy strategy, coupled with chemometric analysis, provides a highly accurate and rapid method for protein toxin detection and classification.
• This approach shows significant potential for enhancing public health protection through the development of advanced toxin detection devices.
• The established models offer alternative pathways for the creation of rapid, field-deployable toxin identification systems.