Machine learning-assisted 3D SERS chip with acoustic enrichment for high-accuracy diagnosis of respiratory viruses and emerging pathogens

Abstract

Outbreaks of SARS-CoV-2, first investigated as an unknown pathogen, have reflected the severe threat that pathogen X poses to public health and social security. Early and precise diagnosis and classification of infectious respiratory diseases with similar symptoms are essential for the risk assessment of public health or epidemiological investigations. Current technologies are limited to detect known viruses, leading to false negatives for novel or mutated pathogens. Here, we propose an ML-assisted SERS strategy for screening various types of respiratory viruses and potential pathogen X in cases with similar infectious symptoms. A label-free 3D plasmonic Au-PS SERS chip was designed to amplify the Raman signal over 103-fold compared to a conventional Au substrate. An ensemble ML model was developed to analyze SERS data for effectively distinguishing between healthy individuals, SARS-CoV-2, RSV, and influenza A and B, as well as identifying newly emerging pathogens. Our experiments demonstrated that the ensemble model integrated with SERS spectra achieved a remarkable classification accuracy of 100%. Notably, the model exhibited excellent performance in detecting mixed viral infections and simulated pathogen X, with a reliable detection range of viral concentrations from 5 × 102 to 106 PFU/mL under acoustic enrichment. This approach holds significant promise for the early screening and detection of emerging and known respiratory pathogens.