Label-free detection of respiratory virus co-infections: integrating surface-enhanced Raman scattering with deep learning

Concurrent or sequential infections by multiple respiratory viruses pose significant challenges for diagnosis, treatment, and disease management, necessitating accurate and rapid identification methods. Here, we introduce a label-free diagnostic platform that combines surface-enhanced Raman scattering (SERS) with deep learning to enable rapid, quantitative detection of respiratory virus co-infections. Utilizing highly sensitive silica-coated silver nanorod array substrates, we collected over 1.2 million SERS spectra from 11 viruses, nine two-virus mixtures, and four three-virus mixtures at varying concentrations in saliva. A novel deep learning model, MultiplexCR, was developed to simultaneously identify virus species and quantify their concentrations from the SERS spectra. The model achieved an exceptional classification accuracy of 98.6% for virus co-infections and a mean absolute error of 0.028 for concentration regression. In blind testing, MultiplexCR consistently demonstrated high accuracy and precise concentration predictions. This SERS-MultiplexCR platform performs the entire detection process within 15 minutes, highlighting its potential for rapid, pointof- care diagnostics in infection detection, as well as broader applications in food safety and environmental monitoring.