Rapid, On-Site SARS-CoV‑2 Variant Detection and Differentiation Using GLAD-Pristine Silver Nanorod Arrays and Machine...

Area of Science:

• Nanotechnology and Spectroscopy
• Biomedical Engineering
• Infectious Disease Diagnostics

Background:

• The emergence of SARS-CoV-2 variants necessitates advanced diagnostic tools for effective pandemic control.
• Current diagnostic methods may struggle with precise differentiation of closely related viral strains.
• Surface-enhanced Raman scattering (SERS) offers high sensitivity for molecular detection.

Purpose of the Study:

• To develop and validate a machine learning (ML)-integrated SERS methodology for accurate detection and differentiation of SARS-CoV-2 variants and subvariants.
• To assess the performance of Support Vector Machine (SVM) and Bidirectional Long Short-Term Memory (BiLSTM) models in classifying viral strains using SERS data.
• To establish a sensitive and specific platform for on-site COVID-19 variant identification.

Main Methods:

• Fabrication of silver nanorod substrates using glancing angle deposition for SERS analysis.
• Detection and differentiation of SARS-CoV-2 wildtype, kappa, delta, omicron variants, and omicron subvariants (BA.1, BA.2, BA.5, XBB).
• Application of SVM and BiLSTM machine learning models to SERS spectral data from 122 clinical nasopharyngeal swab samples.

Main Results:

• The SERS platform demonstrated a detection limit of approximately 100 pfu/mL for SARS-CoV-2 variants.
• The SVM classifier achieved 88.79% accuracy for variant classification and 95.83% for omicron subvariant classification on the validation set.
• The BiLSTM model achieved 85.98% accuracy for variant classification on the validation set, with both models showing promising results on blind test sets.

Conclusions:

• The integrated ML-SERS approach provides a sensitive and specific method for differentiating SARS-CoV-2 variants and subvariants in clinical samples.
• This methodology holds potential for on-site disease prediction and enhanced pandemic management strategies.
• Further validation and optimization could lead to widespread clinical application for rapid variant identification.