Machine Learning Integrated Near-Infrared Surface-Enhanced Raman Spectroscopy for Accurate Strain-Level Virus Identification

Na Zhang, Ziyang Wang, Xingjun Wang, Gabriel A. Vázquez-Lizardi, P. Varela, Dorleta Jiménez de Aberasturi, David Sánchez, Néstor Perea‐López, Samuel J. Lin, RyeAnne Ricker, Edgar Dimitrov, Alexander Sredenschek, Kalana D. Halanayake, Yin‐Ting Yeh, Julian A. Mintz, Jiarong Ye, Xiaolei Huang, Huaguang Lu, Elodie Ghedin, Danielle Reifsnyder Hickey, Luis M. Liz‐Marzán, Shengxi Huang, Mauricio TerronesarXiv 2025 · DOI · arXiv

Abstract

Strain-level identification of viruses is critical for effective public health responses to potential outbreaks, yet current diagnostic methods often lack the necessary speed or sensitivity. Surface-enhanced Raman spectroscopy (SERS) offers great potential for fast and precise virus clarification through unique vibrational fingerprints of biological components. However, existing protocols typically operate outside of the tissue's transparent near-infrared (NIR) window, and are further limited by the intrinsic complexity of clinical viral samples, which complicates spectral analysis and recognition. Here, we report an artificial intelligence (AI)-empowered NIR-SERS platform that integrates machine learning with a rationally designed hybrid substrate: gold nanostars (AuNSt) coupled with gold-coated carbon nanotube arrays (AuCNT). This architecture generates highly localized plasmonic hot spots resonant tuned to NIR excitation, as confirmed by electron energy-loss spectroscopy (EELS), enabling effective signal amplification from viral components. Our system and protocols provide accurate classification of respiratory viruses, including influenza viruses and coronaviruses, not only at the type and subtype levels, but also the more challenging strain level. This approach overcomes the plasmonic mismatch in conventional SERS and the lack of generalizability in AI-driven diagnostics. It shows promise for enhancing rapid virus detection and identification of novel strains and outbreak response capabilities, thus potentially addressing critical challenges in global public health preparedness.

Introduction

Viruses mutate rapidly, leading to new strains with varying levels of transmissibility and resistance to treatment. While PCR is the gold standard for identification, it is slow and requires specialized personnel. Rapid antigen tests are faster but often lack the sensitivity to detect viruses in early stages or the resolution to distinguish between closely related strains.

Surface-Enhanced Raman Spectroscopy (SERS) has long been a candidate for a "label-free" alternative because it provides a unique "vibrational fingerprint" of biological molecules. However, two major hurdles remained:

  1. Fluorescence Interference: Most SERS systems use visible light, which causes biological samples to fluoresce, drowning out the weak Raman signal.
  2. Plasmonic Mismatch: Most SERS substrates are optimized for visible light. Moving to the Near-Infrared (NIR) range—the "tissue diagnostic window" where fluorescence is lower—usually results in a significant loss of signal enhancement.

Methods

The authors tackled these hurdles through a "rationally designed" hybrid substrate and a custom machine learning pipeline.

Substrate Fabrication

The team created a hierarchical structure consisting of Gold Nanostars (AuNSt) deposited onto Gold-coated Carbon Nanotube (AuCNT) arrays. The CNTs provide a high-surface-area scaffold for virus capture, while the sharp tips of the nanostars act as "lightning rods," creating intense electromagnetic "hot spots" when coupled with the underlying gold film.

Schematic of the AuNSt-AuCNT hybrid substrate fabrication process.

NIR Optimization and AI Integration

Using Electron Energy-Loss Spectroscopy (EELS) and FDTD simulations, the team proved that this specific geometry shifts the plasmonic resonance into the NIR range (785 nm), perfectly matching their excitation laser. To handle the complex, high-dimensional spectral data, they employed a 1D Convolutional Neural Network (CNN). Unlike traditional methods, the CNN learns to extract discriminative features directly from the raw spectra without manual preprocessing.

Datasets

The platform was validated using a diverse set of respiratory viruses, including Influenza A (subtypes H1N1 and H3N2), Influenza B, and Avian Infectious Bronchitis Virus (IBV). The model was specifically tested on multiple strains of H1N1 (e.g., California/07/2009 vs. Michigan/45/2015) and H3N2.

Results

The "aha moment" lies in the coupling between the nanostar tips and the gold-coated nanotubes. This interaction creates significantly more "hot spots" than either component alone, providing an enhancement factor of ~2000 even in the NIR range.

EELS spectra confirming plasmonic resonance tuned to the NIR window (785 nm).

The 1D CNN achieved F1-scores of 0.98 for virus type, 0.97 for subtype, and up to 0.99 for specific H1N1 strains. The CNN consistently outperformed traditional machine learning models like Support Vector Machines (SVM) and XGBoost, particularly at the challenging strain level. By using feature importance maps, the authors showed that the CNN was focusing on biologically relevant markers, such as RNA nucleic acid signatures (784 cm⁻¹) and lipid membrane contributions (1138 cm⁻¹), rather than random noise.

Discussion

While the results are impressive, several areas for growth remain. Currently, samples are drop-casted and dried; integrating this with microfluidics could allow for "flow-through" sensing in clinical settings. Furthermore, to be useful in a pandemic response, the workflow needs to be adapted for portable, handheld Raman spectrometers. Future work will need to test the system against a broader array of viruses and bacteria, as well as more complex clinical matrices (like raw saliva or nasal swabs) to ensure generalizability.

Reference: Zhang, N., Wang, Z., et al. "Machine Learning Integrated Near-Infrared Surface-Enhanced Raman Spectroscopy for Accurate Strain-Level Virus Identification." (2024).

Summary updated

Abstract

Strain-level identification of viruses is critical for effective public health responses to potential outbreaks, yet current diagnostic methods often lack the necessary speed or sensitivity. Surface-enhanced Raman spectroscopy (SERS) offers great potential for fast and precise virus clarification through unique vibrational fingerprints of biological components. However, existing protocols typically operate outside of the tissue's transparent near-infrared (NIR) window, and are further limited by the intrinsic complexity of clinical viral samples, which complicates spectral analysis and recognition. Here, we report an artificial intelligence (AI)-empowered NIR-SERS platform that integrates machine learning with a rationally designed hybrid substrate: gold nanostars (AuNSt) coupled with gold-coated carbon nanotube arrays (AuCNT). This architecture generates highly localized plasmonic hot spots resonant tuned to NIR excitation, as confirmed by electron energy-loss spectroscopy (EELS), enabling effective signal amplification from viral components. Our system and protocols provide accurate classification of respiratory viruses, including influenza viruses and coronaviruses, not only at the type and subtype levels, but also the more challenging strain level. This approach overcomes the plasmonic mismatch in conventional SERS and the lack of generalizability in AI-driven diagnostics. It shows promise for enhancing rapid virus detection and identification of novel strains and outbreak response capabilities, thus potentially addressing critical challenges in global public health preparedness.
BibTeX
@misc{zhang2025machine,
  author = {Na Zhang and Ziyang Wang and Xingjun Wang and Gabriel A. Vázquez-Lizardi and P. Varela and Dorleta Jiménez de Aberasturi and David Sánchez and Néstor Perea‐López and Samuel J. Lin and RyeAnne Ricker and Edgar Dimitrov and Alexander Sredenschek and Kalana D. Halanayake and Yin‐Ting Yeh and Julian A. Mintz and Jiarong Ye and Xiaolei Huang and Huaguang Lu and Elodie Ghedin and Danielle Reifsnyder Hickey and Luis M. Liz‐Marzán and Shengxi Huang and Mauricio Terrones},
  title = {{Machine Learning Integrated Near-Infrared Surface-Enhanced Raman Spectroscopy for Accurate Strain-Level Virus Identification}},
  year = {2025},
  eprint = {2509.09851},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.09851},
  howpublished = {arXiv},
}
Figure