DeepATsers: a deep learning framework for one-pot SERS biosensor to detect SARS-CoV-2 virus.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Spectroscopy

Background:

• Artificial Intelligence (AI) integration in medical diagnostics enables faster and more accurate disease identification.
• Surface-Enhanced Raman Spectroscopy (SERS) offers a unique fingerprint-like spectral data for molecular detection.

Purpose of the Study:

• To develop and validate a novel deep learning model, DeepATsers, for rapid and accurate detection of COVID-19 using SERS biosensors.
• To assess the model's ability to identify specific SARS-CoV-2 proteins and generalize across different datasets.

Main Methods:

• Development of DeepATsers, a deep learning model combining Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN).
• Utilized a one-pot SERS biosensor for experimental spectral data acquisition.
• Applied data augmentation techniques including EMSA, Gaussian-noise, GAN, and K-fold cross-validation to enhance dataset generalization.
• Trained the model using the full spectra wavenumber range for optimal performance.

Main Results:

• DeepATsers accurately identified SARS-CoV-2 proteins (S, N, VLP, Streptavidin, blank) from SERS spectral data.
• Data augmentation, particularly using GAN, increased the dataset size and improved average protein classification accuracy from 0.6000 to 0.9750.
• The model demonstrated superior performance compared to traditional machine learning methods (RF, GBM, SVM, KNN).
• Verification with SARS-CoV-2 Omicron variant spectra confirmed the model's robustness and generalizability.

Conclusions:

• DeepATsers provides a robust, generalized, and generative deep learning framework for analyzing 1D SERS spectral datasets of SARS-CoV-2.
• The AI-driven SERS biosensor approach shows significant promise for rapid and accurate COVID-19 diagnostics.