Extracting True Virus SERS Spectra and Augmenting Data for Improved Virus Classification and Quantification.

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Infectious Disease Diagnostics

Background:

• Surface-enhanced Raman spectroscopy (SERS) offers rapid and sensitive infectious disease diagnostics.
• Biological sample backgrounds obscure true viral signals, complicating detection and data analysis.

Purpose of the Study:

• Develop a deep learning framework to extract pure viral SERS spectra from noisy biological samples.
• Improve SERS-based identification, differentiation, and quantification of respiratory viruses.

Main Methods:

• Utilized dual neural networks to extract true virus SERS spectra and estimate concentration coefficients.
• Augmented spectral datasets across varying virus concentrations in water and saliva.
• Trained XGBoost models on augmented data for classification and concentration prediction.

Main Results:

• Extracted spectra closely matched high-concentration spectra, validating accuracy.
• XGBoost models achieved >92% accuracy and R² > 0.95 on augmented water data.
• Models demonstrated robustness on saliva data, achieving >91% accuracy and R² > 0.9.

Conclusions:

• The deep learning framework successfully extracts clean viral SERS spectra, overcoming background noise.
• This methodology significantly enhances SERS-based diagnostics for infectious diseases.
• The approach advances accurate species identification, differentiation, and quantification.