Accurate virus identification with interpretable Raman signatures by machine learning.

Area of Science:

• Virology
• Spectroscopy
• Machine Learning

Background:

• Rapid virus identification is crucial for public health responses to outbreaks.
• Label-free Raman spectroscopy offers a promising method for fast viral detection.
• Machine learning (ML) can analyze viral Raman spectra for identification.

Purpose of the Study:

• To present a machine learning approach for analyzing Raman spectra of human and avian viruses.
• To develop a convolutional neural network (CNN) classifier for accurate virus identification.
• To interpret the ML model to understand key spectral features used in virus recognition.

Main Methods:

• Utilized a portable virus capture device and label-free Raman spectroscopy.
• Developed and applied a convolutional neural network (CNN) classifier for spectral data analysis.
• Employed a full-gradient algorithm to interpret CNN model responses and identify salient spectral ranges.

Main Results:

• Achieved 99% accuracy in classifying influenza A vs. B viruses.
• Reached 96% accuracy in classifying four subtypes of influenza A.
• Demonstrated 95% accuracy in differentiating enveloped and nonenveloped viruses.
• Attained 99% accuracy in differentiating avian coronavirus (IBV) from other avian viruses.
• Verified that the ML model recognizes key biomolecular signatures (proteins, lipids).

Conclusions:

• The CNN-based ML approach accurately identifies and classifies various viruses using Raman spectroscopy.
• The method provides rapid, high-accuracy viral identification, supporting public health surveillance.
• Interpretation of the ML model highlights the importance of specific biomolecular functional groups in virus recognition.